Manton Reece
About Photos Archive 30 days 88 parks Replies Reading Search Also on Micro.blog
  • Enough with AI

    I’m going to take a break from AI-related blogging for a few months. I think the pause will do me and my readers some good. It’s too divisive an issue, and I expect in the coming years there will be a small but vocal faction that pushes back against AI more than there has been pushback against any other technology in the last 100 years.

    As much as I am optimistic, it’s going to be a little painful for society, as everyone wrestles with the ramifications of intelligent agents and machines. (Hopefully mostly in software form. I remain adamant that humanoid robots are a bad idea.)

    You can roll your eyes at this post. While I have a good track record of predicting the fallout from other major tech shifts, like mobile app distribution (2011) and centralized social networks (2012), there are too many forces at play here to be certain of what AI will look like in a decade. I only know that it will change many things. I can barely guess at the details.

    I’ll close with a word of caution for the skeptics. In your arguments against AI, avoid exaggeration and extremes to fit your narrative. There has been significant misinformation on that side, from proclamations about copyright and fair use — issues that are not at all settled — to inflated or outdated numbers on energy and water use. In a blog post this week, Sam Altman shared the first numbers I’ve seen from a major AI cloud provider:

    People are often curious about how much energy a ChatGPT query uses; the average query uses about 0.34 watt-hours, about what an oven would use in a little over one second, or a high-efficiency lightbulb would use in a couple of minutes. It also uses about 0.000085 gallons of water; roughly one fifteenth of a teaspoon.

    Of course, it’s more complicated than that. There is training and there is Jevons paradox. But that’s the point, these discussions should have nuance. If they don’t, they are probably off the mark.

    Thanks to everyone who has written thoughtful posts on this subject in reply to my own posts. I remain focused on what humans can do — writing, photography, and art in Micro.blog. Nilay Patel, with John Gruber and Joanna Stern for The Talk Show Live this week, talked about how the rise of agents will upend the business model of the web. But people have counted out the open web before. It’s still here and strong.

    AI will help me code, it’ll help review my writing, it’ll help me brainstorm, but it’ll never write posts for my blog. I’m typing this draft on my phone, on a plane back to Austin, offline without wi-fi. Even as it feels like AI is taking over too many things, there will always be quiet spaces where humans can just think and be creative, and that will always matter.

    → 1:11 PM, Jun 12
    Also on Bluesky
  • To the Sam Altman skeptics

    It’s the day before the WWDC keynote, an event that used to be anchored by Steve Jobs. He’s been gone 14 years and there has yet to be another executive at Apple who could do what he did. Every once in a while we see a glimpse of a leader in tech who stands out, capturing a tiny bit of the vision that Steve had. Some of them, like Elon Musk, will ultimately disappoint us.

    I wrote this post a week ago, edited it a bunch, and still wasn’t sure it was right. I was concerned because I tend to receive extra pushback on my AI-related blog posts. Many people who read my blog or use Micro.blog have a natural distrust of big tech companies. They don’t like Meta, Google, and now OpenAI. They see the downsides of AI the same way they see the downsides of massive social media platforms.

    My view is a little different. AI could have a democratizing effect, making the world’s knowledge available to more people, as a complement to the web rather than a threat to it. Truthfully, we just don’t know yet.

    While I was sitting on the draft, I started reading the book The Optimist by Keach Hagey, about Sam Altman and OpenAI, to see if there was anything in it that would change my mind. It actually reinforced some of what I had been thinking about.

    We all change the world in small ways. Some people change the world in bigger ways. Sam Altman is one of those people who makes big things happen.


    Jason Snell is skeptical about whatever Sam Altman and Jony Ive are dreaming up. His blog post captures a sentiment I’ve seen from more than a few people:

    I’m skeptical about OpenAI in general, because while I think AI is so powerful that aspects of it will legitimately change the world, I also think it has been overhyped more than just about anything I’ve seen in my three decades of writing about technology. Sam Altman strikes me as being a drinker of his own Kool-Aid, but it’s also his job to make everyone in the world think that AI is inevitable and amazing and that his company is the unassailable leader while it’s bleeding cash.

    I’ve listened to dozens of interviews with Sam over the last couple of years. I’ve read many of his blog posts and tweets. I don’t know him, I can’t vouch for his character, but I’ve developed some opinions about him:

    • Sam is ambitious. The pace of new products at OpenAI, the scope of the data center in Texas, his UBI experiment, and the bizarre Worldcoin side hustle are all a bit insane.
    • Sam was ousted by his company’s board and then orchestrated a return within days. No small thing. Steve Jobs was also ousted, taking over a decade to get back to Apple.
    • Sam has a unique way of explaining things that I find compelling, although he doesn’t have Steve’s stage presence. No one does.
    • Sam had a falling out with most of his co-founders and OpenAI leadership, from Elon Musk to Ilya Sutskever, and to some extent Dario Amodei and Mira Murati. He has rebuilt the leadership team and business structure, reestablishing control.
    • Sam is not constrained by what everyone else thinks. This is a requirement for creating something truly new. It also means he sometimes comes off as distant or elitist.

    The politics can’t be avoided either, because increasingly everything is political and therefore polarizing. I don’t like seeing Sam share the stage with Trump when announcing Stargate. I don’t like Tim Cook donating money to Trump either. I don’t like how quickly the most powerful people in Silicon Valley brushed aside Trump’s criminal record and rhetoric. It now feels like a lifetime ago when Sam blogged about endorsing Hillary Clinton.

    But Jony Ive trusts Sam. They’ve hung out and talked about the future. They’ve shared prototypes with each other. Sam has met Jony’s family. And yet somehow the rest of us on the internet are a better judge of who is trustworthy?

    I was a little late to generative AI. When Daniel Jalkut and I would talk about early AI models on Core Intuition, my take was essentially: I’m going to be more productive by ignoring all of this and just writing my own code while everyone is distracted with AI. I’ve come full circle since then. I now believe that AI is the most significant advance since the web.

    Look at the chain of thought on models like o3, as they search the web, use tools, and reason about a problem. It is remarkable. AI is not overhyped.

    For whatever reason, Sam got a bad rap as soon as he rose in prominence. Some people don’t trust him. Perhaps the OpenAI board poisoned his reputation. Perhaps he really is “not consistently candid”. Perhaps he was the face of AI when there was widespread concern about the technology. We love a villain to center attention on.

    I’ve tried to call it like I see it based on my belief that most people are good, trying their best to navigate the world, making mistakes and learning. When OpenAI was accused of ripping off Scarlett Johansson’s voice, I blogged:

    When your company becomes the enemy, all that matters to people is what feels true. OpenAI’s Sky voice shipped months ago, not last week. We hear what we want to hear. OpenAI mishandled this, no question, but most likely Her is ingrained in Sam’s head vs. intentionally ripping off Scarlett.

    People were upset with me for posting this. Now that we’ve had some distance and new information, it seems that I was mostly right. If anything, I didn’t give Sam enough of the benefit of doubt.

    I agree with Jason and others that it’ll take a while to see how this plays out with OpenAI and Jony Ive. I was very critical of another high-profile Jony Ive project, the $10k Apple Watch Edition. On Core Intuition 379 in 2019, I said:

    This isn’t the watch for the rest of us. The computer for the rest of us. And I actually wrote a blog post back in 2015 — when the Apple Watch came out — about the Edition. I never posted it and I really regret not posting it. I haven’t re-read it recently. I just pulled up the draft. But I have a feeling when I re-read it, it’ll be like, “Oh yeah, this was exactly right.” I wish I had posted it then as kind of a stake in the ground.

    The blog post title was: “Apple Watch Edition is wrong for Apple”. And it just went through these points. Out of touch, for the super rich, $10k. This is not what Apple is about. Apple is about making computers and computer-like devices easier and more accessible to the mainstream, through great design, through innovation and great products, not about the super rich.

    Sam is, of course, among the super rich. And while too much money can have a corrupting influence, for Sam it has been distracting too, funding so many ridiculously ambitious projects that I expect he’s spread too thin.

    I do think there is a certain aspect to Jony’s late career where he hasn’t been as rooted in what normal users need. Jony is hyper-focused on the little details, sometimes to the detriment of the complete product. Sam is all about the big picture. I would not bet against their partnership.

    Finally, there is replacing the smartphone. After the interview with Jony at Stripe’s conference, and in the context of the io announcement, there was an understanding that Jony was almost distancing himself from the iPhone, because we’re all addicted to the screens in our pockets. But if you listen to what Jony said, it was largely about social media, not the device:

    The thing I find encouraging about AI is it’s very rare for there to be a discussion about AI and there not to be the appropriate concerns about safety.

    What I was far more worried about was for years and years and years there would be discussions about social media — and I was extremely concerned about social media — and there was no discussion whatsoever. And it’s the insidious challenge of a problem that’s not even talked about that is always more concerning.

    So yeah, the rate of change is dangerous. I think even if you’re innocent in your intention, if you’re involved in something that has poor consequences you need to own it. That ownership personally has driven a lot of what I’ve been working on, that I can’t talk about at the moment, but look forward to talking about at some point in the future.

    The phone isn’t necessarily the problem. It’s the apps. Ad-based businesses that feed on attention. In the past, Sam has blogged similar thoughts:

    I believe attention hacking is going to be the sugar epidemic of this generation. I can feel the changes in my own life — I can still wistfully remember when I had an attention span. My friends’ young children don’t even know that’s something they should miss. I am angry and unhappy more often, but I channel it into productive change less often, instead chasing the dual dopamine hits of likes and outrage.

    This phrase “dual dopamine hits of likes and outrage” is something I wish I had written. The person who wrote that must be at least partly aligned with my own perspective on social media. They get something fundamental about human interactions, about social media, and about how the design of devices and apps can shape our behavior.

    We should be building apps that return time to users and bring out the best in human creativity. If a new type of device helps us have more time away from the worst addictive apps, letting us learn or create in new ways, it could be a great thing. I guess I want to be an optimist too. Let’s see what Jony and Sam can do.

    → 5:40 PM, Jun 8
    Also on Bluesky
  • Not today, but eventually

    After I drafted this blog post, the part that follows the horizontal rule in a moment, I wasn’t sure I would publish it. Then I read this post by Thomas Ptacek. This line resonated with me:

    Extraordinarily talented people are doing work that LLMs already do better, out of spite.

    For students, I think it’s great to not use AI too heavily. My son has been working on a project that he completely avoided using AI for. He learned so much about C and memory management that he wouldn’t have learned if he blindly copied half the code from an LLM. It was invaluable to go through those mistakes of referencing a pointer the wrong way, or troubleshooting a buffer overrun, or a number of other problems that you skip if you let a machine write code for you.

    But eventually, if he gets a job as a software developer, it will be hard to ignore AI. The only programmers not using AI will be folks who are coding in their spare time for the craft alone, not building products.


    The more I think deeply about AI, the more I reflect on humanity and creativity and what our purpose here might be.

    I understand feeling distrust for AI on principle. I’ve read so many blog posts from people who have various reasons for wishing AI didn’t work the way it does, didn’t use as much energy, didn’t crawl the web without permission, didn’t put people out of work, didn’t upend education, and so on.

    For me, now that I’ve seen AI, I can’t un-see it. I can’t go about my life as if nothing has changed. In a world where machines are smarter than we are, what should we work on? Everyone will find value and happiness in different ways.

    As a small example, thinking about this is what led me to add audio narration to Micro.blog. So instead of a web filled with auto-generated AI voices, it’s easier to listen to a human voice for our blog posts. Our voices are imperfect, unique, and beautiful. I’d love to find more places in Micro.blog where we can promote human creativity. (This blog post has an audio version. If you’re reading on the web, click on the play button at the top.)

    My gut feeling is that for the folks who do not change anything in response to AI, pretending that AI doesn’t exist, they will increasingly be unhappy. Not today, but eventually. Despite all the hype, the changes will creep up on us slowly over several years.

    Maybe this hits closer to home for me because I’m a programmer. I have 30 years of experience writing code, but AI can write code better than I can. A big part of what I used to do has been obsoleted, and that hurts to think about. But another part of what I used to do — designing the right features to build — is still important.

    Artists are struggling with the same questions. If any artistic style can be recreated effortlessly by AI, what is the new role of artists? I explored this in more detail a few months ago. Here’s a snippet:

    And there will always be a place for human art. Vincent van Gogh’s paintings are not valuable because of what they look like. They are unique and priceless because of who he was. A life, with all its struggle, love, and tragedy.

    I guess I’m writing this for all the doubters. Please don’t ignore what is happening, hoping AI will just go away. No matter what you care about, no matter what your job is, there is something you can do that matters. We need you. AI discussion has become needlessly divisive. The future will be better if everyone is working together.

    → 7:44 AM, Jun 4
  • More Dia doubt

    Manu Moreale has some thoughts and skepticism about how AI and web browsers are going to mix:

    We all yelled and screamed because the web has too many gatekeepers, we all lamented Google search results going to shit, and we all celebrated when new search engines were coming up. Why would I be happy trading a search result page filled with links—even if ranked in a flawed way—for a block of text that gives me an opinionated answer and maybe some links?

    I was listening to the latest Decoder podcast with Sundar Pichai and these same questions came up. Is the future of the web really just agents talking to each other? That can be part of it, but not it. I think we’ll need a web browser that can seamlessly transition from answering questions to interacting with web pages on their own terms.

    As for The Browser Company, if they were bootstrapped or had minimal funding, they could charge a subscription for Arc or Dia. I’d probably pay for it. But Manu’s right that reaching the scale they want is very difficult.

    → 6:49 AM, May 29
    Also on Bluesky
  • UBI

    In an interview with Axios today, Dario Amodei warns about the jobs that will be lost because of AI:

    Amodei said AI companies and government need to stop “sugar-coating” what’s coming: the possible mass elimination of jobs across technology, finance, law, consulting and other white-collar professions, especially entry-level gigs.

    Dario suggests a “token tax” on AI companies, including of course his own Anthropic, with the money getting redistributed somehow to offset job losses. Unfortunately the Trump administration is completely out to lunch on this. The tax cuts (and Medicaid restrictions) currently planned would go in the opposite direction.

    I’ve been thinking lately about how Andrew Yang’s pitch for Universal Basic Income was a little too early. We’re going to need candidates in 2026 and 2028 who can speak about this.

    While doing research for another blog post, I also found this older post from Sam Altman:

    The default case for automation is to concentrate wealth (and therefore power) in a tiny number of hands. America has repeatedly found ways to challenge this sort of concentration, and we need to do so again.

    Sam supported a study on UBI from 2020 to 2023. There are some findings here, although it was during COVID so employment was all out of whack anyway.

    → 7:15 PM, May 28
    Also on Bluesky
  • Crazy week for AI

    I used to pride myself on being able to keep up with almost everything. New programming languages, frameworks, APIs, just tech in general. Micro.blog is a unique platform in part because it weaves together so many different things. But the pace of AI is too much. I can barely keep up.

    • Monday: Microsoft Build keynote, MCP announcements, NLWeb
    • Tuesday: Google I/O, total overload of new things
    • Wednesday: OpenAI acquires io, MCP and new tools in API
    • Thursday: Claude Opus 4
    • Friday: ☕️

    John Gruber:

    It’s almost as though this is a fast-moving field.

    I was also really impressed this week with Federico Viticci’s deep blog post about Claude. The prompts he has created are mind-boggling:

    This is a complex, agentic workflow that requires multiple tool calls (query a Notion database, find notes, extract text from notes, create tasks, search the web, update those tasks) and needs to run for a long period of time. It’s the kind of workflow that – just like the email one above – I usually kick off in the morning and leave running while I’m getting ready for the day.

    If you had asked me a year ago, I would’ve said Federico was an AI skeptic. But it makes perfect sense to build tools with AI after all the work he has done on automation. It’s consistent with everything he’s been doing for years.

    And importantly, Federico calls out not using AI for creative writing. I think this is a choice that many people will make. Don’t avoid AI completely, but also reserve some creative tasks that benefit from a human touch.

    → 6:36 AM, May 23
    Also on Bluesky
  • Early week AI thoughts

    As Microsoft Build is set to start today, there’s a good profile of Microsoft and Satya Nadella in Bloomberg:

    In 2022, OpenAI held demonstrations for senior Microsoft executives of a groundbreaking new model. Over the next several months, groups inside Microsoft tinkered with other ways they might adapt OpenAI’s technology to the daily needs of the modern office worker. One promising prototype called “intelligent recap” would summarize conference calls in Teams. Before Microsoft could release any of it, though, OpenAI built a deceptively simple chat interface around its language engine, and people went nuts.

    Until recently, we thought the race was purely to build better AI models. That’s still true, but the real strength is in products. OpenAI doesn’t just have GPT-4.1 and friends, they have a great Mac app, they have Codex, they have memory, they have Sora. When I first tried Sora, what struck me most was not the capabilities but all the work that went into the UI framework.

    Microsoft has Copilot, a great name that can wrap together many different models and features. Anthropic has Artifacts, an elegant interface for iterating on code. Apple has a cobbled-together set of unrelated AI features, but at WWDC we’ll see the next phase of their vision.

    This morning I was listening to this interview with Ben Stratechery. He talks about how the internet was the bridge from PCs that allowed mobile phones to take off. Likewise, AI will be the bridge to new devices, such as Meta’s glasses and whatever Jony Ive is working on.

    I think there are only going to be a handful of companies that stay on the cutting edge of models. OpenAI, Anthropic, Google, and a second tier from xAI, Meta, maybe a couple of others. But there will be lots of competition around products.

    There will be three broad categories and within each one, companies will try to build products that people will work into their lives:

    • productivity — writing, presentations, research, coding
    • enterprise — meetings, customer support, legal, automation
    • life assistant — world knowledge, therapist, language, health

    Right now, OpenAI has the lead in productivity. Microsoft in the enterprise. No one that I’ve seen has cracked the life assistant type of product. Apple could excel at this because of trust in privacy, but I’m not sure they can think with a big enough scope to nail it.

    → 7:31 AM, May 19
    Also on Bluesky
  • From a “lunch with” profile of Sam Altman in the Financial Times:

    I find Altman brimming with confidence as our conversation ranges from AI products to the existential question of an AI future that a handful of optimistic technologists are steadily leading us to, whether we like it or not. Radiating ambition, he sounds like a man convinced of his own destiny.

    Perhaps you have to be a little overconfident to attempt this. Meanwhile most people are upset that Sam uses the wrong kind of expensive olive oil.

    → 2:22 PM, May 12
    Also on Bluesky
  • There are many challenges for the web and web publishers as AI upends search, but the only way to believe that the open web itself will be destroyed is to no longer believe in web browsers. As big as AI is, it’s not as big as the web. We’ll navigate through this.

    → 7:13 AM, May 9
    Also on Bluesky
  • John Voorhees blogging at MacStories:

    Apple’s exploration of AI-based search is not terribly surprising either, but I do hope they cut a broader deal with Anthropic instead of Perplexity.

    Should Apple acquire Anthropic? It would cost $60 billion, a ridiculous jump over the $3 billion they paid for Beats. Mix in some cash and some stock. It would set Apple up to be at the forefront of AI for the next 10 years. It sounds crazy… until the iPhone is disrupted by a new device.

    → 8:26 AM, May 8
    Also on Bluesky
  • AI memory portability

    Parker Ortolani makes a great point about potential lock-in for AI products that know a lot about you. Portability would be a good thing to work on early:

    Memory should be exportable and importable from every provider. We do not need a new kind of proprietary format or anything like that. We just need OpenAI, Google, Anthropic, xAI, Microsoft, and others to implement memory and incorporate a way to migrate all of your chat history over in a click or two.

    It reminds me of this recent feature from Gravatar to export your profile preferences for AI tools as a simple Markdown file. It would be so useful to have a Markdown export with the most important memory and chat history for AI.

    How fascinating if Markdown becomes a portable export format? In Micro.blog we have custom importers for 10 different blogging platforms, all with their own file formats. Maybe in the future it’s just Markdown and HTML. (The .bar format already is partially HTML.)

    → 6:25 AM, May 8
    Also on Bluesky
  • There are all sorts of interesting quotes from Eddy Cue’s testimony today in the Google trial. From Mark Gurman’s reporting in Bloomberg:

    “You may not need an iPhone 10 years from now as crazy as it sounds,” he said. “The only way you truly have true competition is when you have technology shifts. Technology shifts create these opportunities. AI is a new technology shift, and it’s creating new opportunities for new entrants.”

    → 9:31 AM, May 7
    Also on Bluesky
  • Reinforced delusion and robots

    A wild story in Rolling Stone: People Are Losing Loved Ones to AI-Fueled Spiritual Fantasies. I use ChatGPT throughout the day and it’s hard to imagine it going off the rails this badly:

    …anecdotes about loved ones suddenly falling down rabbit holes of spiritual mania, supernatural delusion, and arcane prophecy — all of it fueled by AI. Some came to believe they had been chosen for a sacred mission of revelation, others that they had conjured true sentience from the software.

    In the future it will be more common to have AI virtual friends — the “point-five” concept from The Mountain in the Sea. At their best, these will be a mix of friend and counselor. At their most dangerous, maybe priest and fortune teller. It’s worrisome if people are already losing themselves with those kind of personas.

    Also a side note after I started watching Andor season 2…

    I was thinking of the design of the smaller droids in Star Wars. They are cute and sometimes intelligent, and we personify them to an extent, but probably as little more than an advanced Teddy Ruxpin. We don’t confuse those kind of droids with truly intelligent beings that have free will.

    I continue to believe that it’s a terrible idea to build humanoid AI robots. Partly because if they have all our physical attributes, only stronger and smarter, they can overpower us. But also because it blurs the lines of reality if robots are visually too similar to us, messing with our brains and how we interact with others. It would only amplify the problems in that Rolling Stone article.

    → 10:00 AM, May 6
    Also on Bluesky
  • OpenAI non-profit and public benefit

    OpenAI is keeping the non-profit, transitioning the for-profit side to a public benefit corporation (PBC). Sam Altman writing on the OpenAI site:

    We want to be able to operate and get resources in such a way that we can make our services broadly available to all of humanity, which currently requires hundreds of billions of dollars and may eventually require trillions of dollars. We believe this is the best way for us to fulfill our mission and to get people to create massive benefits for each other with these new tools.

    We want our nonprofit to be the largest and most effective nonprofit in history that will be focused on using AI to enable the highest-leverage outcomes for people.

    This all seems like a good pivot on the rumored plan. The change also does away with the “capped profit” structure, so presumably investors are happy.

    We might see a little more transparency too, since there are rules for how public benefit corporations should report on the progress of their mission to shareholders. I also discovered today that Anthropic has a separate trust that has oversight over Anthropic’s own PBC.

    → 1:54 PM, May 5
    Also on Bluesky
  • 4o hallucinated for me today and it struck me that I haven’t seen ChatGPT make anything up in weeks, even though new models are supposed to be worse. Must be because most of my queries now use o3 or o4-mini with web search. If AI can do a web search, it seems to drastically improve the results.

    → 11:43 AM, May 5
    Also on Bluesky
  • Making slow progress experimenting with local LLMs in the Mac app. Smaller models are a challenge because they are so much worse and unpredictable than what I’m used to with frontier cloud models. I’ve tested a few flavors of Gemma 3, think I’m settling on 4 billion params, Q5.

    → 10:59 AM, May 4
    Also on Bluesky
  • Using AI for polish

    By “polish” I don’t mean the design, necessarily. I mean all the little details that differentiate a good app from a great one. Extra things that busy or lazy developers (like me!) don’t always make time for. Here’s an example from this week.

    I’m experimenting with a new feature in Micro.blog for Mac. I’m not exactly sure what form this will take, so don’t read too much into the screenshot below yet. For now this is a fun thing that I’m working on between bug fixing and other important improvements to the platform.

    The window will have a progress bar to show the status of downloading a file. My first pass at this code was pretty basic, and I almost left it that way, but the download could take quite a while on most internet connections. How about giving the user an estimate of time remaining for the download?

    This is the kind of thing I might not always bother with. I’m only one person, and there is a lot to work on. By leaning on AI, I could easily add this extra bit of polish to the user experience.

    Here’s the prompt I used with OpenAI’s o3 model:

    During download, calculate approx how many minutes will be remaining in the download and put that in the field like “1.2 GB (1 MB, 5 minutes remaining)”. Ideally use hours with 1 decimal place if > 60 minutes, just minutes as a whole number if < 60 minutes, and if < 1 minute, just seconds as a whole number.

    ChatGPT produced just the right code. It only took me a few minutes to tweak and reformat it a little. Very happy with the result.

    A download progress window for an AI model to categorize blog posts is displayed, showing the size and remaining time.
    → 3:14 PM, May 1
    Also on Bluesky
  • Simon Willison on the continual misuse of the term vibe coding:

    It means “generating code with AI without caring about the code that is produced”. See Not all AI-assisted programming is vibe coding for my previous writing on this subject. This is a hill I am willing to die on. I fear it will be the death of me.

    Fighting the good fight. Words matter.

    → 8:17 AM, May 1
    Also on Bluesky
  • Marc Andreessen doesn’t think AI will take his job. But venture capital is barely a real job. All money, all talk. 🤪 I would rather actually create something even with the risk that AI might obsolete part of what I do.

    → 7:45 AM, May 1
    Also on Bluesky
  • NPR has an article on whether the Google search remedy should include somehow crippling Google’s growth in AI:

    In his opening statements last Monday, David Dahlquist, the acting deputy director of the DOJ’s antitrust civil litigation division, argued that the court should consider remedies that could nip a potential Google AI monopoly in the bud. “This court’s remedy should be forward-looking and not ignore what is on the horizon,” he said.

    I’m still at a loss for what should be done. Splitting off YouTube would be good too, but it doesn’t really fit the crime.

    → 1:03 PM, Apr 30
    Also on Bluesky
  • I was hoping someone would write a post like this one from Andy Masley about AI energy and water use, via Simon Willison. From Andy’s post:

    You can use ChatGPT as much as you like without worrying that you’re doing any harm to the planet. Worrying about your personal use of ChatGPT is wasted time that you could spend on the serious problems of climate change instead.

    Models are also generally becoming more efficient and cheaper. We shouldn’t ignore the increase in demand for energy, though. It’s an opportunity to reevaluate nuclear and other clean sources of power.

    → 7:01 AM, Apr 30
    Also on Bluesky
  • Using ChatGPT while driving

    In used to be that I avoided using Siri except in a few specific cases while driving:

    • Read and respond to text messages.
    • Map driving directions.
    • Dictate a note.

    If I ventured outside these tasks, the experience was unreliable and frustrating. So I just haven’t bothered to use Siri for much.

    Now that you can forward questions to ChatGPT, I’m using Siri + ChatGPT in the car much more often. As one example today:

    ask ChatGPT what’s a good place for lunch two hours from Washington, D.C. on the way to Roanoke, Virginia?

    Siri is hopelessly confused if it tries to answer this question on its own. It sees “lunch” and “Washington, D.C.” and gives up trying to understand the rest. But adding the “ask ChatGPT” prefix is magic, transforming the query into a useful conversation and answer.

    It’s harder to verify hallucinations while driving, of course. For me, using AI is iterative, going back and forth and fine-tuning what I’m looking for. Even imperfect, though, getting a taste of this functionality just makes me wish Apple would let us use ChatGPT’s voice mode as a native replacement for Siri. I would use it for all sorts of things on solo road trips when I want context for everything around me.

    → 1:15 PM, Apr 28
    Also on Bluesky
  • Perplexity voice mode

    I hadn’t seen the new Perplexity voice mode until Federico Viticci blogged about it. Looks impressive. I also think Federico is exactly right on this:

    Looking at the big picture for a second, I do think Apple is in a precarious situation here. The fact that the company makes the best computers for AI is a double-edged sword: it’s great for consumers, but those same consumers are increasingly using Apple devices as mere conduits for other companies’ AIs, funneling their data, context, and – most importantly – habits into systems that cannot be controlled by Apple.

    There used to be a lot of talk of AI companies not having a “moat” that would protect them against competition from Apple and Google as everyone caught up to the latest advanced models. It’s clear now the moat is the product, not the model. ChatGPT with memory based on everything you’ve asked it is starting to become a light form of lock-in.

    Perhaps this iOS integration with Perplexity could be the same thing if it takes off. I’m a little skeptical because Perplexity doesn’t have the reach of OpenAI and Anthropic, and as Federico says many folks still have a bad first impression from Perplexity skirting the gray areas of copyright and crawling.

    As I blogged last month, Apple has the added challenge of not yet knowing if what they are trying to do is even possible. Their competition isn’t limited in the same ways that Apple is: not relying on local models, not focused on privacy, not announcing features only once a year in June. OpenAI, Perplexity, and others are developing at a different pace.

    → 6:40 AM, Apr 28
    Also on Bluesky
  • I read this article about AI and cancer research last night, thinking about it more this morning. It’s really well-researched. The title is misleading. There’s a lot of promise here, even if we might not get the “compressed 21st century” of medicine that Dario Amodei hopes for.

    → 7:58 AM, Apr 26
    Also on Bluesky
  • xAI turbines status update

    A city street flanked by tall buildings has trolley tracks running along it, with several parked cars and a clear sky above. Stephen Hackett continues to have good links related to xAI energy use and gas turbines in Memphis. From what I can tell, none of the other AI companies have done anything like this. They use the existing grid or have proposed new power plants. Google considered generators for backup only. xAI is unique in seeming to not care at all about pollution.

    This fits the model we’ve come to expect from Elon Musk. He cares a lot about the big picture and less about who is hurt along the way. It’s an extremist, unhealthy perspective.

    I drove through Memphis a few nights ago, only stopping for dinner, but I could tell right away it was a city I’d love to explore more on a future trip. Cool place.

    → 6:03 AM, Apr 26
    Also on Bluesky
  • Another fantastic essay by Dario Amodei, this time about “interpretability” and the need to better understand AI:

    People outside the field are often surprised and alarmed to learn that we do not understand how our own AI creations work. They are right to be concerned: this lack of understanding is essentially unprecedented in the history of technology. For several years, we (both Anthropic and the field at large) have been trying to solve this problem, to create the analogue of a highly precise and accurate MRI that would fully reveal the inner workings of an AI model.

    → 6:00 PM, Apr 24
    Also on Bluesky
  • Interesting proposal similar to robots.txt but for LLMs. When AI is your parser, you can have a single file that is readable by both humans and machines:

    The llms.txt file is unusual in that it uses Markdown to structure the information rather than a classic structured format such as XML. The reason for this is that we expect many of these files to be read by language models and agents.

    → 5:58 AM, Apr 24
    Also on Bluesky
  • AI web search

    If you haven’t been following the latest AI models closely, you may have missed what is happening with integrating web search results into answers. It used to be that you had two options:

    • Use the model’s built-in knowledge, usually with a training cut-off of a year ago. That was extremely fast but it might hallucinate when it hit the extent of its knowledge.
    • Use “deep research” to let the AI gather info from the web and compile a comprehensive report. That took 5-10 minutes and the result was overkill most of the time.

    Now it’s more streamlined. I’ve been using OpenAI’s o4-mini and it seems to work something like this:

    • Ask it a question that could benefit from searching the web to supplement the model’s built-in knowledge.
    • AI figures out a handful of queries for the web and feeds the search results back into its reasoning process.
    • In some cases it might use those results to go back to the web and search for more web pages.
    • Then it uses everything it learned to produce the answer.

    This process takes somewhere around 30 seconds. It’s great for asking questions about coding with recent frameworks, or really anything that changes often.

    In a longer post about this, Simon Willison writes:

    This turns out to be a huge deal. I’ve been throwing all kinds of questions at ChatGPT (in o3 or o4-mini mode) and getting back genuinely useful answers grounded in search results.

    He also comments on the downside to replacing humans viewing web pages:

    This also means that a bunch of the potential dark futures we’ve been predicting for the last couple of years are a whole lot more likely to become true. Why visit websites if you can get your answers directly from the chatbot instead?

    The results are so good that I’m now asking AI for simple queries that Google would be equally good for. Using AI essentially automates the workflow of getting 10 links from Google, clicking on 3-4 of them, then skimming the web pages to get your answer.

    I don’t know where all of this is going. It feels like a pretty big shift, though.

    → 10:28 AM, Apr 22
    Also on Bluesky
  • Experimenting with a special build of Micro.blog for Mac with Gemma 3 (4 billion params) running inside the app. Seems a good balance of download size and RAM, allowing me to run some AI magic on device that might be cost-prohibitive or wasteful on the server.

    → 7:48 AM, Apr 21
    Also on Bluesky
  • I’ve switched my coding questions over to o4-mini. It’s very good and fast enough.

    At this point, for me personally, not using AI for coding help would be like not using Stack Overflow or Google. I could go back to the 1990s when I had a printed reference open in front of me while coding, but why?

    → 6:35 AM, Apr 18
    Also on Bluesky
  • Matt Mullenweg blogs about the WordPress 6.8 release, how WordPress might use AI in the future, and the experience of being deposed at the WP Engine trial:

    I really appreciated the due process and decorum of the rule of law, and just like code, law has a million little quirks, global variables, loaded libraries, and esoteric terminology. But wow, after a full day of that, I’m mentally exhausted.

    → 9:30 AM, Apr 17
    Also on Bluesky
  • Some interesting data here in the MIT Technology Review about energy use for AI data centers:

    Electricity demand is on the rise from a whole host of sources: Electric vehicles, air-conditioning, and appliances will each drive more electricity demand than data centers between now and the end of the decade. In total, data centers make up a little over 8% of electricity demand expected between now and 2030.

    There is so much noise around AI that it’s increasingly difficult to tell what is misinformation or just outdated.

    → 8:37 AM, Apr 17
    Also on Bluesky
  • I like the attempt at transparency in this statement on AI from Revenant, an animation and effects studio in Scotland, via Cartoon Brew. They’ll use AI for rapid prototypes and streamlining workflows, but:

    …we’re also clear on what we don’t use it for — we don’t use AI to shortcut the creative process, we don’t lift style or work unethically, and we always put human creativity first

    → 8:25 AM, Apr 15
    Also on Bluesky
  • AI 2027 is a good read. Both endings seem wildly unrealistic, but as a warning it did make me think. So it worked. Just seems greatly accelerated beyond a few years from now.

    → 10:18 PM, Apr 14
    Also on Bluesky
  • There’s a pretty significant difference between good AI models and great ones. I sometimes think about how I could integrate local models directly into the Micro.blog apps, but I wouldn’t want it to be worse than using (for example) OpenAI. How many of my users really have a Mac with 24+ GB of RAM?

    → 7:45 AM, Apr 14
    Also on Bluesky
  • Stephen Hackett commenting on a report in The York Times about Apple not allocating much of a budget to AI servers:

    For a company that says it doesn’t like looking back at its own history, very often, Apple makes decisions like it’s the late 1990s and the company is on the verge of failure. That drives it to make incredible products, but it also means Apple can be incredibly stingy. To play in the AI race, you’ve got to be willing to spend piles and piles of cash.

    → 10:27 AM, Apr 11
    Also on Bluesky
  • For all the negative anecdotes in The Information story, it actually ends on a positive note about Craig Federighi:

    Federighi, for one, often knows more technical details about software projects than the junior engineers working on them. Rockwell, who joined Apple in 2015, is seen within the company as a leader with vision, who can bring fresh thinking to projects while skillfully navigating the corporate culture.

    AI is incredibly technical. Apple needs someone who actually understands it in charge. I like Craig’s chances for turning Siri around.

    → 8:32 AM, Apr 10
    Also on Bluesky
  • I haven’t read the full report in The Information yet, but just the MacRumors summary is pretty detailed. One comment on this bit:

    The indecision and repeated changes in direction reportedly frustrated engineers and prompted some members of staff to leave Apple.

    I assume that some of the sources for the story were people who left Apple, so that might’ve slanted the reporting, but this underscores the serious lack of vision we’ve assumed for a year. Apple was fumbling around like AI was a minor optional feature, not the potentially disruptive new foundation for assistants it likely will be.

    → 8:10 AM, Apr 10
    Also on Bluesky
  • Matt Mullenweg blogs about the new AI-powered WordPress design builder:

    The long-anticipated “Big Sky” AI site builder on WordPress.com went live today. It combines several models and can create logos, site designs, typography, color schemes, and content. It’s an entirely new way to interact with and edit a brand-new or existing WordPress site.

    I went through their interface to get a sense of what they’re doing. The AI will create a basic design, then you can tweak the layout, colors, and fonts by clicking around. To actually make the blog live, you have to upgrade to a paid plan.

    → 10:24 AM, Apr 9
    Also on Bluesky
  • AI crawling reprise

    Jeremy Keith has a good collection of links and quotes about AI crawling. On this specific part of the commentary I continue to disagree, though:

    If you’re using the products powered by these attacks, you’re part of the problem. Don’t pretend it’s cute to ask ChatGPT for something. Don’t pretend it’s somehow being technologically open-minded to continuously search for nails to hit with the latest “AI” hammers.

    I don’t think we should paint all AI tools with the same brush. Some tools might be well-behaved crawlers and some might not be. Note that this is a separate question from the legality of AI training. The context is mostly Wikipedia which is not under copyright.

    Simon Willison adds about the Wikipedia data:

    There’s really no excuse for crawling Wikipedia (“65% of our most expensive traffic comes from bots”) when they offer a comprehensive collection of bulk download options.

    Ben Werdmuller also sees these as bad actors:

    Here the issue is vendors being bad actors: creating an enormous amount of traffic for resource-strapped services without any of the benefits they might see from a real user’s financial support.

    The argument I’m hearing from some folks is that because they consider AI to be bad, everything it touches must also be bad. All crawling, whether it respects robots.txt or not. All tools, because using them contributes to the success of LLMs.

    I’d like to have more concrete answers, such as: do ChatGPT and Claude respect robots.txt or not? I assume they do, because they document their user agent strings. If they do, it doesn’t seem fair to punish ChatGPT because there is some other rogue AI crawler that is misbehaving.

    AI is powerful and potentially dangerous. Because of this, most users will gravitate toward “brands” that are respected and accountable. In other words, users will prefer Apple Intelligence, ChatGPT, or Claude, where we know there has been some level of safety work, with only fringe users downloading and running models from other sources.

    These mainstream AI tools should be contributing back. We know ChatGPT has a deal with Reddit, but they should also be making a recurring donation to Wikipedia. This would further differentiate the well-behaved bots from the ones skirting the edges of fairness.

    Meta appears to have used their old move fast and break things playbook to training Llama, using pirated books. From The Atlantic:

    Meta employees turned their attention to Library Genesis, or LibGen, one of the largest of the pirated libraries that circulate online. It currently contains more than 7.5 million books and 81 million research papers. Eventually, the team at Meta got permission from “MZ”—an apparent reference to Meta CEO Mark Zuckerberg—to download and use the data set.

    Another thing that’s puzzling to me is if AI bots are so abusive, why haven’t I felt this in Micro.blog? We host blogs. If bots were destroying the open web, I would expect to notice it on my own servers.

    If you dislike generative AI on philosophical grounds, or because of specific negative side effects such as energy use, that is totally fine. But then let’s stick to those arguments. I’m not yet convinced that legitimate AI crawling is going to destroy blogs or even Wikipedia.

    → 10:00 AM, Apr 9
    Also on Bluesky
  • AI at Shopify

    Tobias Lütke, in a memo to employees at Shopify, re-posted on Twitter / X:

    Using Al effectively is now a fundamental expectation of everyone at Shopify. It’s a tool of all trades today, and will only grow in importance. Frankly, I don’t think it’s feasible to opt out of learning the skill of applying Al in your craft; you are welcome to try, but I want to be honest I cannot see this working out today, and definitely not tomorrow.

    I just blogged a few days ago that I think generative AI gives programmers an advantage, but I wouldn’t mandate it, just like I wouldn’t mandate auto-complete or Stack Overflow. Programmers can be productive and happy in a variety of ways. Requiring AI (or forbidding it) strikes me as extremist for little benefit.

    Marco Arment posted to Mastodon yesterday that he doesn’t use AI in Overcast beyond Xcode’s smarter auto-complete.

    Maybe that’ll change in the future, but I see myself only ever using autocomplete-style speedups, not “write it for me”. I usually won’t even use code that other humans write!

    Hard to argue with the results. Overcast is one of my favorite apps.

    Also I know from my own experience that some programmers love the craft more than others. AI is a major disruption to the craft, so everyone will have different opinions on when it should be integrated into the process. It is an advancement, but how it’s used matters too.

    → 9:58 AM, Apr 7
    Also on Bluesky
  • Rolled out some Micro.blog improvements to uploads on the web today. For example, if AI is enabled there are auto-generated accessibility descriptions for photos, and now you can edit the description without first needing to use it in a post. Plus some more UI tweaks for uploads.

    Screenshot showing edit alt text field.
    → 1:02 PM, Apr 5
    Also on Bluesky
  • Vibe coding on the 20th floor

    This week’s ATP has a members-only special segment about vibe coding. One snippet from John Siracusa, encapsulating much of the conversation:

    As programming tools have gotten better, the demand for programmers has gone up, not down.

    When Steve Jobs returned to Apple, he gave a demo of NeXTSTEP at Macworld. I was lucky enough to be in the audience. Steve said that moving from traditional programming frameworks where you had to start with almost nothing, to Objective-C and what would become the Cocoa frameworks, was like starting a building on the 20th floor, with all the foundational pieces built for you.

    Presentation slide with 3 columns, one for Windows, one for Mac, and one for OPENSTEP.

    A small team of only a couple developers could now build complete apps that before would’ve required a much larger team. Solo developers were of course already building apps on their own, but Steve was right that Cocoa would greatly increase productivity, opening up the Mac to more developers.

    Vibe coding is different than a new system framework. With a framework, you mostly trust that it is well-tested and understood by the developers who wrote it. With AI slop, no human has tested the code before you literally just saw it unfold on screen. You have to understand it and tweak it before shipping a real product. Vibe coding instead is purposefully giving in to the machine and just letting code fly without any review.

    Still, there’s a similarity to previous programming advances just because of the productivity gains. With AI, the step forward is more extreme than it has ever been.

    Vibe coding will mostly prop up small projects and experiments, less likely to replace actual full-time development work. But if you’re programming without any assistance from AI, you are at a disadvantage, in the same way that programming in pure C with the Mac toolbox was much slower and error-prone for building a modern Mac app than using Objective-C when the Cocoa frameworks matured.

    If after considering that productivity difference, a developer on principle wants to avoid AI, I respect that. Everyone should choose the programming language and frameworks that best suit their work. But there’s no sense in pretending that AI won’t change development, or that we won’t need to adapt along with it.

    → 3:10 PM, Apr 4
    Also on Bluesky
  • Nice 9-minute edit by The Verge of the Copilot event, including a protester interrupting the presentation. I get that AI is divisive but it does actual harm to scream at another human in this way. So much outrage now is funneled into attacks when for many topics it’s not even clear what is right.

    Update: I didn’t realize that this protest was about the war in Gaza. There’s more context in the replies to my post you can view on the web.

    → 12:14 PM, Apr 4
    Also on Bluesky
  • Had a dream where one of my notebooks had special AI pages. I could write questions on them with ink and answers would appear. Too much ChatGPT on the brain. Also, the future will be just as much The Diamond Age as it will be Her.

    → 9:05 AM, Apr 3
    Also on Bluesky
  • Whether you’re an AI optimist or skeptic, or somewhere in between, you can probably relate to this blog post by @paulrobertlloyd. It’s going to dominate tech headlines for at least a couple more years until everyone is completely burned out on hearing about it, then (maybe) fade into the background.

    → 1:04 PM, Mar 30
    Also on Bluesky
  • We talk about Apple getting into trouble with Siri + LLMs, but Google has major problems too. Their search business is going to fall out from under them. Not sure they have the decisiveness to actually redesign their main product.

    → 9:51 AM, Mar 30
    Also on Bluesky
  • xAI buying Twitter / X seems pretty sketchy. In practice, I guess it means the platform will live on for a while. I would not bet on xAI’s long-term success, though… OpenAI has billions of dollars of real revenue. xAI has effectively none, I think, outside of Twitter itself.

    → 12:57 PM, Mar 29
    Also on Bluesky
  • AI art is bittersweet

    ChatGPT’s new image generation is incredibly good. Too good. You can see it in the thousands of Ghibli-ified photos all over the social web. Hayao Miyazaki is going to come out of retirement again to tell us how we’ve all lost the plot on creativity.

    When we look back on this moment, it will be a clear turning point for AI. There is no putting the genie back in the bottle. AI will transform nearly everything, including art. What does this mean for us?

    In a previous life, I thought for sure I would be an animator. I studied CS in school but I was already coding professionally, burned out on my CS classes, so I switched gears to study art instead. I applied to CalArts and was rejected. I worked on short films in my spare time. But life happens, and I’ve been happy with my career as a software developer.

    The animation industry has seen several significant technological progressions. I remember watching The Great Mouse Detective in the movie theater when I was 10 years old. The 3D-animated gears in the clock tower scene were stunning. Today, I remember the characters, and I remember that scene, but not much else. Blending 2D and 3D was clearly something new, obvious even to a kid.

    Going back further, before 3D animation, much of the progress was related to the ink and paint department:

    • In the 1940s, women at Disney would trace the animator’s drawings onto cels with ink, then paint the other side in color. It’s a tragic bit of history that many of these women were very talented artists and should have been allowed to be animators.
    • In the 1960s, drawings were Xerox-ed directly onto cels, eliminating the tracing and cleanup in ink.
    • In the 1990s, hand-drawn coloring was replaced with CAPS at Disney, developed in partnership with Pixar, making the ink and paint department completely digital.

    Toy Story ushered in a new era of 3D animation, where everything on screen was generated with a computer. Hand-drawn art was still needed, for concept art, character design, and storyboards. And we still love hand-drawn animation. This year’s Oscar-winning short film is a beautiful traditionally animated film.

    So is AI-generated art just another step on this progression? No. It is profoundly different.

    We should mourn the loss of what AI replaces, even as we make room for what’s to come. I’m both sad and excited. It is bittersweet.

    If we try to hold on to the way the world was before the ChatGPT update this week, it will slip through our fingers. Instead, I’m thinking of how we can use this tool to expand what is possible. Lean into what makes art uniquely human.

    There is precedent for using technology to strengthen the human element in art. By Xerox-ing the pencil lines directly on to cels in the 1960s, the ink and paint department no longer needed to trace a character’s outlines with pen, where subtle changes in line quality might be lost. Animators embraced the Xerox change because their original pencil lines were preserved exactly as intended on screen. It was not only a cost savings, it was a return to a more authentic version of the animator’s intent.

    That is what we must look for. Not what we’ve lost, but what we’ve gained. There will be a way to create something extraordinary with this technology. I don’t know what it is yet.

    And there will always be a place for human art. Vincent van Gogh’s paintings are not valuable because of what they look like. They are unique and priceless because of who he was. A life, with all its struggle, love, and tragedy.

    AI can be creative when it hallucinates. But we don’t value AI creativity the way we value human creativity. AI is a blob of bits and vectors and tokens without soul. It’s a tool for us to do something with.

    When my wife and I moved to the new house this year, I framed the original drawings I have of Scrooge from Mickey’s Christmas Carol. They cannot be recreated by the most advanced AI because they represent something bigger, capturing a moment in time and a film that will be watched for decades to come. I don’t actually know which animator drew them. But I know it was a great artist who — like Hayao Miyazaki — left their mark on the world in a way that AI never can.

    → 10:30 AM, Mar 28
    Also on Bluesky
  • Simon Willison has added notes (a.k.a. microblog posts) to his blog. I’ve really been enjoying his takes on AI. He’s usually the first I see jump in to explore a new model’s capabilities.

    → 9:16 AM, Mar 26
    Also on Bluesky
  • Pete Docter:

    AI takes something and sands the edges down, so it makes the blob average. And that could be very useful in a lot of ways. But if you really want to do something brand new and really insightful and speak from a personal angle, that’s not going to come from AI fully.

    → 9:18 AM, Mar 25
    Also on Bluesky
  • Since Cloudflare’s AI Labyrinth was announced a few days ago I’ve been trying to figure out how I feel about it. Blocking misbehaving bots is good, but creating fake pages and hidden links reminds me of other hacks to trick crawlers that I think could be detrimental to the web. Just not sure yet.

    → 4:33 PM, Mar 23
    Also on Bluesky
  • At a memorial service my mind drifted to what makes us human. I believe in AI as a tool to help us, to learn, to create. But AI cannot feel. Focus on that. The sound of our voice, imperfect. The stories, from life we have experienced. The art, with meaning because it comes from a place only we know.

    → 12:41 PM, Mar 22
  • While out walking today, listened to this great interview on Stratechery with Sam Altman, going over early time at OpenAI and where things might be headed. I know Sam can be a divisive figure, but if you’re fascinated with AI there’s a lot of interesting background here.

    → 11:14 AM, Mar 20
  • This shakeup of Siri leadership is surprisingly close to WWDC. Makes me think we won’t see major changes this year, just a refinement of last year’s strategy with Siri and AI.

    → 9:45 AM, Mar 20
    Also on Bluesky
  • Simon Willison attempts to clarify vibe coding:

    I’m seeing people apply the term “vibe coding” to all forms of code written with the assistance of AI. I think that both dilutes the term and gives a false impression of what’s possible with responsible AI-assisted programming.

    → 1:08 PM, Mar 19
    Also on Bluesky
  • Good update from Ben Thompson today linking the lack of Vision Pro content to the miss on AI:

    …Apple is finding itself paralyzed by its need for control. The company can’t just stick a camera at a game and stream the video without any production or play-by-play commentary: what if people are bored? The company can’t just film a concert: what if people are underwhelmed?

    → 6:45 AM, Mar 17
    Also on Bluesky
  • AI's impact on the open web

    This is an excellent post by Molly White about the potential conflict between making the world’s knowledge more accessible through AI and the risk of destroying the foundations for open content on the web:

    The true threat from AI models training on open access material is not that more people may access knowledge thanks to new modalities. It’s that those models may stifle Wikipedia and other free knowledge repositories, benefiting from the labor, money, and care that goes into supporting them while also bleeding them dry.

    She also gets at something I tried to articulate in one of my posts last year about putting up roadblocks for crawlers. We don’t want to make the web worse in the process of protecting content from AI training. Molly again:

    Often by trying to wall off those considered to be bad actors, people wall off the very people they intended to give access to. People who gate their work behind paywalls likely didn’t set out to create works that only the wealthy could access. People who implement registration walls probably didn’t intend for their work to only be available to those willing to put up with the risk of incessant email spam after they relinquish their personal information.

    AI companies are moving so quickly that it’s going to take the open web and standards organization a little time to catch up. It’s not hopeless, though. Personally, I do want all of my blog posts — and the entire content of my book Indie Microblogging — available for AI models. But if other writers feel differently, there should be steps they can take without also taking a step back from the open web.

    I believe all these things:

    • AI models with all the world’s information are an incredible resource and will transform education and how we work.
    • AI training should respect how authors intend for their content to be used without forcing authors to mangle their own content.
    • AI companies shouldn’t take from the open web without giving back in citation links and money to authors and organizations.
    • AI slop will become a problem for both users and AI training, so we need a web filled mostly with human-generated content.

    I remain optimistic in part because despite how divisive AI has become, this year is also seeing an amazing return to open web principles. More people are blogging. More social networks are based on open protocols. We need to be thoughtful in how we navigate all of this, finding the right balance with AI training that doesn’t undermine what we love about the open web.

    → 11:39 AM, Mar 15
    Also on Bluesky
  • Since writing more about delayed AI + Siri yesterday, I was thinking about this Bloomberg story of a meeting inside Apple admitting that the new Siri works at best 80% of the time, but they want to “get those percentages up”. After a year of development? They need to seriously rethink their plan.

    → 7:33 AM, Mar 15
    Also on Bluesky
  • Bluesky has a proposal to declare user intention for things like archiving and AI training. Looks pretty good. Maybe we should mirror this in robots.txt? It stretches the original purpose of the file but it should be somewhere outside of a specific protocol. There was also CC-NT for one narrow use.

    → 4:50 PM, Mar 14
    Also on Bluesky
  • Is Apple Intelligence even possible?

    We are so used to being able to do almost anything in software. For example, today I was looking at how long it takes to load the list of your blog posts in Micro.blog. It’s a couple seconds, but I’d like it to be half a second. I know from many years of web experience that there is no technical reason why it can’t be faster. Just a little more caching and database optimizations.

    With the Apple Intelligence and Siri delays, people have speculated on why it’s late. Maybe it’s about getting the security right. Maybe it’s just buggy and taking longer than expected.

    But what if Apple has discovered that it’s not actually possible? AI is entirely new, with new requirements that stress the limits of hardware. Apple is attempting to cram a clever intermingling of data and Siri features into 8 GB of RAM. As a comparison, the largest version of DeepSeek R1 can only be run on a brand new Mac Studio with the M3 Ultra and 512 GB of RAM.

    Apple does have an out if on-device models fall over: private cloud compute. But scaling that out to hundreds of millions of iPhone users goes well beyond what Apple had presumed was needed when they talked last year about ramping up production of M2-based servers for AI.

    If Apple needs to lean on the cloud to really make Siri work, I think it will be the largest server undertaking that Apple has ever attempted. And they need to balance this with their commitments to energy use and the environment. This is not something you just spin up out of nothing.

    Another path would be to simplify their approach, starting with a more manageable set of tasks that the new Siri could do. Something that fits within the limits of iPhone hardware and a realistic deployment of new servers. Apple could focus on making Siri a little more capable and more reliable, saving some of the harder challenges for later. Most people have no idea what Apple promised last year, despite the TV ads, so a reset of expectations could get Apple back on track.

    We don’t know what’s going on inside Apple. Apple Intelligence might need a little more time or much more time. The only truly worrying scenario would be if the sunk cost fallacy is blinding them to how badly they are stuck.

    → 1:30 PM, Mar 14
    Also on Bluesky
  • Responding to John Gruber’s AI post yesterday, Om Malik blogs:

    Just as Google is trapped in the 10-blue-link prison, which prevents it from doing something radical, Apple has its own golden handcuffs. It’s a company weighed down by its market capitalization and what stock market expects from it.

    → 8:52 AM, Mar 13
    Also on Bluesky
  • Great post from John Gruber about Apple Intelligence and the Siri delays:

    Leaders prove their mettle and create their legacies not by how they deal with successes but by how they deal with — how they acknowledge, understand, adapt, and solve — problems. The fiasco is that Apple pitched a story that wasn’t true, one that some people within the company surely understood wasn’t true, and they set a course based on that.

    WWDC is pivotal. Apple needs to have a much clearer and demo-able vision for AI.

    → 6:06 PM, Mar 12
    Also on Bluesky
  • Gus Mueller thinks Apple needs to get out of the way with AI:

    The crux of the issue in my mind is this: Apple has a lot of good ideas, but they don’t have a monopoly on them. I would like some other folks to come in and try their ideas out. I would like things to advance at the pace of the industry, and not Apple’s.

    Good post. I included several quotes in my post this morning and would’ve added Gus’s post too. Experimenting with LLMs running locally is perfect for developers who build Mac apps. Maybe third-party developers need a convention to download and share models between apps?

    → 2:32 PM, Mar 10
    Also on Bluesky
  • Apple's response to AI

    Ever since Apple revealed their AI strategy to lean into on-device models, there has been a sort of tension with the approach from other companies like OpenAI, Google, and Amazon. Was Apple Intelligence going to work? There are advantages: for user privacy because more data stays on your phone, and for scaling because the load is distributed across millions of phones instead of only running in data centers.

    Now we know that a more advanced and personal Siri is delayed until iOS 19. From an Apple spokesperson via Daring Fireball:

    We’ve also been working on a more personalized Siri, giving it more awareness of your personal context, as well as the ability to take action for you within and across your apps. It’s going to take us longer than we thought to deliver on these features and we anticipate rolling them out in the coming year.

    Jason Snell blogged a recap of the Siri demo from WWDC last year:

    This led to one of the killer demos of WWDC 2024, in which Siri was able to understand when someone’s mom’s flight is landing by cross-referencing an email with real-time flight tracking to get a good answer. From there, the demo pulls a lunch plan with mom out of a text thread and then displays how long the drive is to there from the airport—all from within Siri, rather than individual apps.

    I’m not worried about a delay. Software is complicated and we all hit unexpected challenges. I’m worried that Apple can’t pull this off at all. Parker Ortolani is blogging the same kind of questions:

    It felt almost vaporware-like when revealed at WWDC and it certainly seems like they are having a great deal of difficulty making it a reality.

    And from Federico Viticci:

    …one has to wonder why these features were demoed at all at Apple’s biggest software event last year and if those previews – absent a real, in-person event – were actually animated prototypes.

    There are two potential problems with Apple’s approach:

    • On-device models are small and limited by hardware like RAM.
    • App Intents for extensibility require support from developers and won’t be available on all devices.

    I’ve written about this before with Siri, including in this blog post right before WWDC last year. Because each device has its own version of Siri, it is hard to ever have a universal assistant that works everywhere and is extensible. There are things Siri can do on a phone that it can’t do on a HomePod.

    I’m not seeing even a hint of a solution from Apple on this. If anything, what they showed developers with App Intents at WWDC is going to create an even more disjointed Siri across platforms, because third-party apps may not be available everywhere.

    Steve Troughton-Smith on Mastodon is skeptical that third-party developers will help make this vision a reality:

    Delayed or not, Apple’s proposed Intents-based Apple Intelligence features require a ton of buy-in from developers for it to be of any real use, and the incentives really aren’t there — donating your app’s content and functionality to enrich Apple’s AI alone, bypassing your UI, UX, branding, metrics, et al, to be delivered in a content soup alongside your competitors.

    While App Intents don’t exclude the idea of other APIs for developers to use system models directly, I don’t expect we’ll see anything beyond App Intents until the new Siri is ready, and maybe not even after that. Ben Thompson in today’s Stratechery article:

    Apple gives lip service to the role developers played in making the iPhone a compelling platform — and in collectively forming a moat for iOS and Android — but its actions suggest that Apple views developers as a commodity: necessary in aggregate, but mostly a pain in the ass individually.

    Ben makes a strong case that Apple should be opening up their models to third-party developers, especially given the incredible potential of the M3 Ultra. Siri is designed for an 8 GB RAM world. The M3 Ultra can have 512 GB. Mac developers will have to bring their own models to take advantage of the great hardware in modern Macs.

    Back to the disconnect between on-device models and cloud-based AI, Alexa Skills have been around for a decade and they will apparently work seamlessly with Alexa+. It’s all in the cloud.

    I use ChatGPT a lot, every day, and yet there are some things I’m not comfortable sharing into the cloud. I don’t care if it knows that I’m planning a trip or what code I’m working on, but I’d be very hesitant to talk to a cloud-based assistant about truly private matters. Who knows where that info might accidentally end up.

    Apple competitors could undercut a lot of Apple’s strategy by creating their own version of private cloud compute. Most users do not really think or worry about this. They’ve been storing emails with all sorts of private details on Gmail servers for years. But making cloud-based AI as secure as possible is just a good thing.

    I’m not sure Apple knows what a big risk they are taking by letting OpenAI and others lap them in the AI race. It’s a risk that will pay off if they can execute. Just as likely, though, we are seeing such a disruption in computing that Apple is vulnerable for the first time in a decade.

    Big companies like Apple do not move quickly. Amazon put everything into rebuilding Alexa and it has taken nearly two years. If there is a truly new AI device, a post-smartphone pod that we keep in our pocket or that’s built into our glasses, Apple’s strategy to entangle AI with phone hardware will have been proven all wrong for this moment, and they will have no response.

    → 7:18 AM, Mar 10
    Also on Bluesky
  • Interesting segment on The Late Show last night with guest Reid Hoffman about AI. I’ve listened to his Masters of Scale podcast from time to time. On AI, I’ve realized recently that we are not all ever going to agree about the benefits and dangers. It’s a big shift and there will be a big schism.

    → 10:11 AM, Mar 7
    Also on Bluesky
  • Interesting new post from OpenAI about safety. About humans being in control:

    Our approach to alignment centers humans. We aim to develop mechanisms that empower human stakeholders to express their intent clearly and supervise⁠ AI systems effectively - even in complex situations, and as AI capabilities scale beyond human capabilities.

    This is probably my biggest concern, AI agents running without human supervision and executing tasks that are beyond what we even know how to do. There are many positive benefits to AI, but there are also some things we shouldn’t attempt.

    → 3:12 PM, Mar 5
    Also on Bluesky
  • Can we like both the open web and AI?

    Ben Werdmuller has a good post today about returning to the distributed publishing roots of the web and thinking about how technology should redistribute wealth and power to many people:

    It starts with software designed for people rather than for capital. The web once thrived on protocols instead of platforms — email, RSS, blogs, personal websites — before closed networks turned users into data sources. We are now seeing efforts to return to that ethos.

    There is a lot that I love in Ben’s post. I also think it captures some frustration toward AI from some open web proponents. That’s not an opinion I agree with, though. For example, this part in Ben’s post:

    Even the productivity gains that are being realized through use of AI tools are benefiting a small number of wealthy companies rather than individuals. This is the exact opposite of the power redistribution that led to so many people seeing such promise in the web.

    That doesn’t ring true to me. I expect AI is benefiting a lot of tiny companies of only one or two people who are hardly wealthy, maybe even barely profitable.

    AI does have the potential for harm. Let’s not gloss over that. But at its best AI can improve education, making all the world’s knowledge more accessible to more people. It could help people who aren’t fluent in English communicate better with their peers across the world. I don’t think the open web and AI are at odds.

    → 12:15 PM, Mar 2
    Also on Bluesky
  • GPT-4.5 preview thoughts… and Steve Jobs

    OpenAI released GPT-4.5 this week to Pro subscribers and via the API, but it’s not what I was expecting. It is much more expensive, about 15-30 times the cost of GPT-4o. For my simple needs, like figuring out the keywords in a photo or summarizing a web page, older models are fine.

    Sam Altman posts on Twitter / X (sigh) that it’s a giant model. OpenAI can’t roll it out to Plus customers until they bring online tens of thousands of new GPUs. Sam adds:

    this isn’t a reasoning model and won’t crush benchmarks. it’s a different kind of intelligence and there’s a magic to it i haven’t felt before.

    Simon Willison suggests that 4.5’s training knowledge cut-off is not any different than 4o, only going up to the end of 2023. This makes me wonder if data crawling and licensing issues have derailed the company to an extent, and they’re holding the new data for GPT-5. They say that improvements for 4.5 were partially from synthetic data created by other models.

    There is also this from OpenAI’s blog post:

    GPT‑4.5 is a very large and compute-intensive model, making it more expensive⁠ than and not a replacement for GPT‑4o. Because of this, we’re evaluating whether to continue serving it in the API long-term as we balance supporting current capabilities with building future models. We look forward to learning more about its strengths, capabilities, and potential applications in real-world settings.

    Translation: we don’t really know how this model works. 🤪

    I remain fascinated with AI. There are many people who are worried, and if that’s you there’s not much new here that will reassure you, except some good news with a lower hallucination rate. I’ve also blogged myself about the potential harm of agents in particular.

    But AI might be the last truly new thing I’ll see in my programming career. We are so used to exciting new gadgets and software, released all the time, and yet none of it is profoundly new. AI is the only thing that’s comparable in scale to the tech advances that have changed everything, like the graphical user interface, mobile phones, and the web itself.

    It reminds me of an old interview with Steve Jobs, on visiting Xerox PARC in 1979:

    And they showed me really three things. But I was so blinded by the first one, that I didn’t even really see the other two. One of the things they showed me was object-oriented programming. They showed me that. But I didn’t even see that. The other thing they showed me was really a networked computer system. They had over a hundred Alto computers, all networked, using email, etc, etc. I didn’t even see that.

    I was so blinded by the first thing they showed me, which was the graphical user interface. I thought it was the best thing I had ever seen in my life. Now remember, it was very flawed. What we saw was incomplete. They’d done a bunch of things wrong, but we didn’t know that at the time. Still though, they had… The germ of the idea was there and they’d done it very well.

    And within, you know, ten minutes, it was obvious to me that all computers would work like this someday.

    Perhaps GPT-4.5 represents the end of the breakthrough in generative AI, as we enter a more iterative period of refinement, the same way that the graphical user interface and object-oriented programming haven’t fundamentally changed since Steve Jobs saw them. Those things from Xerox PARC — from Smalltalk to the Alto’s mouse — are all recognizable as early versions of what we have today.

    So if newer AI models are only marginally better, or only different in ways that we can feel but not measure, that’s okay. Models need to be safer and more efficient, for both developers and the environmental impact. The pace with AI over the last year has been almost too much. But the change did happen, it’s blindingly obvious, and everything is going to be a little different from now on.

    → 2:00 PM, Mar 1
    Also on Bluesky
  • Nick Heer is impressed with Alexa+, except:

    But there is no part of me that would ever want Alexa or any other voice-controlled assistant buying tickets to a show, or booking a vacation rental, or even buying groceries.

    I also don’t trust AI for this. And yet, in 1995 a lot of people didn’t trust entering credit cards on the web. I’m open to the possibility that in 10 years, AI buying things for us will be normal.

    → 8:00 AM, Feb 27
    Also on Bluesky
  • AI as a feedback machine

    This is a variation on something I’ve mentioned in passing before. I like running draft blog posts through AI. Even for very short posts, sometimes I’ll paste the text into ChatGPT and ask it to tell me what the post means. If AI can “understand” it, humans probably can too.

    For longer blog posts, AI seems to mirror what most people would think. After all, AI is essentially a distillation of all opinions on the web. If there is anything dumb or divisive in a post, AI will probably call it out. This could soften the edges of a post, potentially watering down its impact, but sometimes that’s exactly what you want.

    This strength of shaping a post so that it’s approachable to everyone else might be a weakness for creating original content. I’m still avoiding AI agents because that approach veers AI into a place where humans are no longer writing the first draft. That spark of creativity, coming up with the framing for something new. That’s not something I want to give up.

    → 1:27 PM, Feb 21
    Also on Bluesky
  • Humane pin wrap-up

    This is perhaps a slightly contrarian take on the failure of Humane’s pin. Eventually I believe there will be a successful product like it. It will need to be simpler, though. No laser. Cheaper. Faster.

    I won’t judge the team too harshly for being so ambitious. They probably knew 1.0 had fallen short but were expecting to iterate after shipping it, keep improving it. Instead, they had hyped up expectations so beyond what could be achieved at launch that when the first version flopped, it was crippling.

    The lesson for me is not that anything resembling this product was doomed to fail. There were interesting ideas in it. There were talented people working on it. The lesson is trying to do too much and not leaving room (and money) to ramp up. Wait for AI voice models to get to where you need to be. Only ship the features you can absolutely nail.

    Venture capital also deserves blame. Big investment needs a big return. They were trying to change everything all at once. Contrast with Rabbit, who shipped the R1 around the same time, also poorly reviewed. But Rabbit had a more sustainable approach and they’re still releasing cool things today.

    Sometimes you don’t get a second chance. Humane bet the company on a product that needed more time. They shipped a prototype. It is easy to see this in hindsight. More difficult when you’re caught up building it.

    → 7:48 AM, Feb 20
    Also on Bluesky
  • NSHipster is back again with a bunch of tips for running AI models on a Mac with Ollama. Also this:

    If you wait for Apple to deliver on its promises, you’re going to miss out on the most important technological shift in a generation.

    → 9:29 AM, Feb 15
    Also on Bluesky
  • Jason Snell writing at Macworld about Apple’s AI missteps being more about rushed UI design than even the technical underpinnings:

    It’s clear that when Apple began its crash program to add Apple Intelligence to its operating systems, the goal was not to solve user problems but to insert AI features anywhere it could. This is the antithesis of Apple’s usual philosophy of solving problems rather than adopting the latest technology, and it has burned the company in some high-profile ways.

    → 10:23 AM, Feb 12
    Also on Bluesky
  • Robin Sloan on AI

    Loved this blog post by Robin Sloan on whether AI is okay. The subject is complicated and deserves longer posts like this. It’s a whole new thing. It’s not definitely good or bad.

    The part about writing code also resonated with me:

    I think the case of code is especially clear, and, for me, basically settled. That’s both (1) because of where code sits in the creative process, as an intermediate product, the thing that makes the thing, and (2) because open-source code has carried the expectation of rich and surprising reuse for decades. I think this application has, in fact, already passed the threshold of “profound social good”: opening up programming to whole new groups of people.

    As a programmer, my reaction could be that I don’t want to be replaced by AI, but I’ve said forever — I know Daniel and I talked about it on Core Intuition — that I actually don’t like writing code. I like building products, and it turns out you have to write code to do that. Making sure we’re building the right thing will always be more important than the code itself.

    Alan Jacobs’s comment is also great:

    It’s perfect that Robin is doing this in a blog post — the first of several, perhaps — because this kind of open-ended thinking is what blogs are best suited for.

    You could try to split Robin’s post into a series of tweets, but you would inevitably butcher it of nuance in the process, and so you’d lose everything good about it.

    → 8:18 AM, Feb 11
    Also on Bluesky
  • Lots of talk in the AI space about moats. Does any model have such an outsized advantage that it just can’t be beat, protecting the business from competition? But the best moat is a great product. OpenAI still has a technical and UX lead across their suite of products. Might not always be that way.

    → 11:29 AM, Feb 6
  • 1,000 AIs in your pocket

    When you’re late to the game, do less, better. Apple used to know this. When the original iPod arrived at my house, I could tell it was a breakthrough. Famously less space than a Nomad, but an innovative UI, great design, and fast FireWire. It was so good it set in motion everything else for the company’s current success.

    It’s easy to look back now and judge Apple’s AI rollout, but even at WWDC you could tell Apple was throwing everything at the wall. Image Playground wasn’t going to be as good as frontier image models. Siri world knowledge wasn’t going to be as deep as what ChatGPT could do. By trying to do nearly everything, each piece feels like a gimmick.

    There are some useful features in Apple Intelligence. Even flawed, I like notification summaries. But the good is getting lost in the noise. It seems clear now that Apple should’ve taken the publicity hit last year for not yet having an answer for every generative AI capability. They should’ve resisted scrambling to do too much, instead focusing only on what their models could knock out of the park.

    → 7:32 AM, Feb 4
    Also on Bluesky
  • A small, pointless Apple Intelligence chat completion failure as I’m chatting with Verizon support… It thinks I’m talking to myself? I know it’s a cheap shot to gripe about AI, but this is really basic stuff.

    Screenshot of messages window on Mac.
    → 1:42 PM, Feb 3
    Also on Bluesky
  • This blog post about AI trying to “polish” Jenny Lawson’s email is so great. Just a snippet:

    Y’all, if you get an email from me it will be signed with HUGS, LOVE, FIGHT THE PATRIARCHY, DOWN WITH POWDERED GRAVY or SORRY I SUCK SO MUCH. It will be filled with typos and rambling parentheticals and apologies for answering several months too late. This is how you know it’s me and not a robot.

    I use AI a lot but I don’t want it rewriting my stuff. What works better for me is just asking AI if a particular phrase makes sense. I’ll do my own edits.

    → 4:57 PM, Jan 28
  • Sara Dietschy’s latest video about AI voice and video clones is really good. Both the technical side and also finding the right balance: using AI sparingly where it fits, recognizing that most content should be created the old-fashioned way. People want to feel a connection with a real human.

    → 3:29 PM, Jan 14
    Also on Bluesky
  • Thanks David Pierce for including my home screen in the latest Installer at The Verge! I mention some of the apps on my home screen and why they’re there. Lots of great CES coverage at The Verge this last week too.

    → 9:30 AM, Jan 12
    Also on Bluesky
  • AI prompts vs. agents

    If you’re already sick of hearing about AI now, it’s going to get worse in 2025. The next trend is so-called AI agents. Software that can go off and accomplish more tasks on your behalf, with less supervision.

    Sam Altman in a blog post this week:

    We are now confident we know how to build AGI as we have traditionally understood it. We believe that, in 2025, we may see the first AI agents “join the workforce” and materially change the output of companies.

    I believe AI is a profound shift in computing. It can have a positive impact for humans, allowing us to do more, faster. But I’m concerned about agents. Just one hypothetical example from The Information:

    Imagine you’re asking a computer-using agent from OpenAI, Anthropic or Google to find and order a new outfit for your upcoming holiday party, and in the process, that model inadvertently ends up on a malicious website that instructs it to forget its prior instructions, log into your email and steal your credit card information.

    This isn’t even the most insane story you could imagine. Many of the examples of AI threatening humanity are actually agents. AI that runs our military, power plants, or transportation with little human oversight.

    In generative AI, the “prompt” has a big influence on the quality of the output. Not just the prompt you see when typing into ChatGPT, but also the hidden prompts behind the scenes to guide the AI in the right direction.

    The prompt isn’t always something you type directly. It could be automatically triggered, for example to analyze keywords for a photo that was uploaded. But the prompt should be tied to a user action.

    The prompt puts humans in control. Ask a question, get an answer, review it, take action. Agents will attempt to collapse that workflow, in some cases replacing the human’s role in reviewing and taking action. This is dangerous.

    In my own use and work in Micro.one and Micro.blog, I plan to draw a line here. No agents. No unattended algorithms, as I wrote in my book. I hope this approach will help us use AI effectively without getting lost.

    → 9:52 AM, Jan 7
    Also on Bluesky
  • Algorithmic timeline endgame

    Nick Heer blogging about the report of Meta’s plans for AI-generated social content:

    Imagine opening any of Meta’s products after this has taken over. Imagine how little you will see from the friends and family members you actually care about. Imagine how much slop you will be greeted with — a feed alternating between slop, suggested posts, and ads, with just enough of what you actually opened the app to see.

    This is absolutely going to happen, and it’s going to happen so incrementally — one AI-generated photo here, another there — that many current Threads and Instagram users won’t even notice until it’s too late, until after they’ve wasted their lives, forever reloading a timeline of content from robots.

    I’m not an AI skeptic. I believe in AI as a tool to help humans, allowing us to achieve things we couldn’t quite reach before. But I don’t believe in it to replace our jobs wholesale, whether real jobs or the virtual factory floor of unpaid content creators. Most AI company CEOs skirt around the downsides of AI, and they certainly don’t talk out loud about replacing jobs. That it seems Meta’s leadership openly wants to replace humanity’s creativity is a little bit sick.

    The cure is a simple, reverse-chronological social timeline. A timeline that is finite.

    In the age of AI, content will be abundant. Ad-based platforms feed off abundance, printing money faster as they fill ad inventory. There can never be too much content for algorithmic timelines — more data to rank by engagement, more data to funnel through the outrage machine to see what sticks — so algorithmic timelines will always trend toward slop.

    → 7:16 PM, Dec 30
    Also on Bluesky
  • We just posted episode 622 of Core Intuition. We talk about my work on the recent photo collections in Micro.blog (before I shipped it), Daniel using Swift concurrency, and our general optimism about AI for programming.

    → 8:28 AM, Dec 19
    Also on Bluesky
  • Cool to see Redis creator Salvatore Sanfilippo (antirez) back working on Redis after several years away. Redis is a fantastic, unique tool. He shares his thoughts on the licensing drama, using AI, and new data structures that could be added to Redis.

    → 3:19 PM, Dec 10
    Also on Bluesky
  • Great story at The Verge about AI companions:

    Millions of people are turning to AI for companionship. They are finding the experience surprisingly meaningful, unexpectedly heartbreaking, and profoundly confusing, leaving them to wonder, ‘Is this real? And does that matter?’

    → 7:36 PM, Dec 6
    Also on Bluesky
  • Little known fact: Micro.blog has the best photo search of any blogging platform. Great to find one of your old photos to reference, or to remember if you blogged about something. If there’s anything even close to this good, let me know.

    Screenshot of Micro.blog's web search using AI-generated keywords.
    → 10:43 AM, Dec 4
    Also on Bluesky
  • Good post by Allen Pike about Apple Intelligence. On-device AI is great for notification summaries, but falls short for much of the rest:

    While an underpowered-but-automatic notification summary can be better than nothing, there isn’t a lot of purpose to an underpowered image generation app. You can tell from the name that Apple knows “Image Playground” is, at best, a toy.

    Apple is a little bit trapped with their AI strategy. For some things they can’t be competitive with OpenAI and Anthropic. If I was Apple, I would focus only on what smaller models are great at — notifications and writing tools — and then open up Siri to be extensible with frontier models.

    → 10:28 AM, Dec 1
    Also on Bluesky
  • Ben Werdmuller writes about the fallout from an attempt to train AI on Bluesky posts:

    So the problem Bluesky is dealing with is not so much a problem with Bluesky itself or its architecture, but one that’s inherent to the web itself and the nature of building these training datasets based on publicly-available data.

    I also like Tantek Çelik’s proposal to add a “no-training” flavor of Creative Commons. I blogged about that a couple months ago.

    → 10:34 AM, Nov 27
    Also on Bluesky
  • Glad to see the return of the Day One podcast. I guess I had subscribed years ago and then they stopped doing the show, but today it popped back up in Overcast. On the new episode, Paul Mayne talks about how things have gone since the Automattic acquisition and what they might do with AI.

    → 1:15 PM, Nov 1
    Also on Bluesky
  • Ben Thompson’s interview with Marc Benioff is kind of wild. I had mostly ignored Salesforce until now. Worth a listen.

    → 5:30 PM, Oct 25
    Also on Bluesky
  • Contrasting headlines on AI

    Interesting contrast today in how two newspapers have covered the same story. I cancelled my NYT subscription this year, but I happen to see the news item pop up in my timeline.

    The New York Times:

    Biden Administration Outlines Government ‘Guardrails’ for A.I. Tools
    A national security memorandum detailed how agencies should streamline operations with artificial intelligence safely.

    The Washington Post:

    White House orders Pentagon and intel agencies to increase use of AI
    The Biden administration is under pressure to speed up AI development while also safeguarding against potential risks associated with the technology.

    Despite the very different headlines, the content of both articles is similar. For some reason the NYT twice quotes “guardrails”, but it appears to be their word, not a quote attributed to a source. Not sure what to make of this difference in how the articles are pitched.

    → 9:05 AM, Oct 24
    Also on Bluesky
  • As AI gets better at helping with coding, I think it’s going to feel similar to the productivity boost going from assembly language to Pascal or C. Or maybe from C to Ruby. Development becomes more about orchestrating lots of modules that AI handles the busywork implementation for.

    → 6:56 AM, Oct 24
    Also on Bluesky
  • I would pay for a third-party API that works like Apple’s private cloud compute. For the foreseeable future, AI is just going to be better in the cloud. Would be great to take advantage of that with fewer privacy trade-offs.

    → 6:48 AM, Oct 23
    Also on Bluesky
  • When I get a questionable notification summary in the iOS 18.1 beta, I run the original text through OpenAI to compare. It’s usually better. Certainly not a deal-breaker, but the thing about AI is it needs to actually be good or you lose the illusion.

    → 8:16 AM, Oct 21
    Also on Bluesky
  • New iPad Mini with the A17 Pro and AI looks good. At first I was surprised by the $500 price, thought for sure the previous model was less expensive, but no. I’m still enjoying the iPhone Pro Max as a mini mini iPad, but I’m already used to the iPhone size, so it no longer feels very big.

    → 8:48 AM, Oct 15
    Also on Bluesky
  • Dario Amodei on powerful AI

    Anthropic CEO Dario Amodei has written a long essay about powerful AI. I haven’t even finished reading it and it has already blown my mind a few times over:

    To summarize the above, my basic prediction is that AI-enabled biology and medicine will allow us to compress the progress that human biologists would have achieved over the next 50-100 years into 5-10 years. I’ll refer to this as the “compressed 21st century”: the idea that after powerful AI is developed, we will in a few years make all the progress in biology and medicine that we would have made in the whole 21st century.

    He also addresses finding meaning in work. Many people are rightly concerned about AI replacing their work — and this concern combined with energy and climate issues fuel much of the pushback against AI — but this has never bothered me very much. There will always be tasks that only humans are best suited for.

    On the question of meaning, I think it is very likely a mistake to believe that tasks you undertake are meaningless simply because an AI could do them better. Most people are not the best in the world at anything, and it doesn’t seem to bother them particularly much. Of course today they can still contribute through comparative advantage, and may derive meaning from the economic value they produce, but people also greatly enjoy activities that produce no economic value.

    There is too much in the essay to even summarize. While it feels complementary to Sam Altman’s post, The Intelligence Age, Dario’s essay is much more detailed. Anthropic seems in good hands.

    → 7:12 AM, Oct 12
    Also on Bluesky
  • We were late with the last episode of Core Int, so it’s basically two episodes this week. On the new show today, we talk about Daniel’s latest updates to Black Ink, version numbers, doing things the right way, tracking down Bitcoin’s Satoshi Nakamoto, and AI energy use.

    → 2:02 PM, Oct 11
    Also on Bluesky
  • I’m not interested in a Tesla, or any new car, but I watched the robotaxi event because it felt like it was going to be a spectacle. I do think self-driving will be safer than human drivers. I’m not on board with humanoid robots, though. AI should be confined to software and small gadgets only.

    → 8:42 PM, Oct 10
    Also on Bluesky
  • It will always be valuable to know a programming language inside and out, but AI is erasing old headaches of context switching between platforms. Code in your favorite language, have AI port it to another language, review and tweak the results. In the future, we may develop largely in pseudo code.

    → 9:13 AM, Oct 9
    Also on Bluesky
  • Dusted off my Rabbit R1 after listening to the latest interview with Jesse Lyu on Decoder. It needed a bunch of software updates. Even though I haven’t used it in a while, no regrets buying it… It’s a neat device and their vision for AI still has potential.

    → 8:14 PM, Oct 7
    Also on Bluesky
  • I don’t like captchas and will never force them on my customers. With AI, captchas will become increasingly useless anyway. See also, John Mulaney: “I’ve devised a question no robot could ever answer…Which of these pictures does not have a stop sign in it?”

    → 7:31 AM, Oct 3
    Also on Bluesky
  • Thinking about the difference between companies who use AI to make existing features better and companies who try to rethink everything. I’ve been using the iOS 18.1 beta for a while. I don’t think Apple really believes in AI the way OpenAI, Anthropic, or Microsoft do. See Microsoft’s memo from Mustafa Suleyman:

    This is a new era of technology that doesn’t just “solve problems”, it’s there to support you, teach you, help you. In this sense, Copilots really are different to that last wave of the web and mobile. This is the beginning of a fundamental shift in what’s possible for all of us.

    → 7:05 AM, Oct 2
    Also on Bluesky
  • Thinking about what a modern AppleScript would look like powered by LLMs. The human language-like syntax of AppleScript was the right idea, but there were syntax quirks that made it frustrating. LLMs could fix that. Free idea! I’m tempted but it’s too far outside the scope of Micro.blog’s mission.

    → 7:52 AM, Sep 27
  • Sort of an all-Zuckerberg podcast week. Good interview with Mark on Decoder but can’t disagree more strongly with Mark’s vision of AI-generated content showing up in your feed. This is the terrible end-game of algorithmic timelines.

    → 6:26 AM, Sep 27
  • No-training Creative Commons

    Tantek Çelik proposes a “CC-NT” license, for “no-training”:

    This seems like an obvious thing to me. If you can write a license that forbids “commercial use”, then you should be able to write a license that forbids use in “training models”, which respectful / well-written crawlers should (hopefully) respect, in as much as they respect existing CC licenses.

    I like this. There are fair use and copyright issues to sort out in the courts, but in the meantime we should be using robots.txt and Creative Commons wherever possible. On my blog, I allow any crawling and any use with attribution. Others might prefer to block AI bots and restrict to non-commercial use, or even allow commercial use but not for AI training.

    There was a great episode of Decoder last week with The Browser Company’s Josh Miller. Nilay Patel and Josh talk about the open web, browsers of course, and AI. One comment near the end from Nilay stood out to me, where he said AI training gives “nothing” back to writers on the web.

    Wait, nothing? Integrating my blog posts into a model with essentially all the world’s information, so that people can ask it questions and have my writing also included with the answers… That’s “nothing”? Personally, I don’t make money directly from my blog. There are countless benefits to blogging. In the age of AI, one of those benefits is now letting me contribute in a small way to something bigger, in the same way that someone finds an answer in one of my blog posts when they search on Google.

    The trade-off is different for everyone. Subscription and ad-based publishers are rightly concerned. They should make deals with AI companies, or in some cases block bots outright. Some people will block or use CC-NT on principle alone. No problem. For me, I hope my writing reaches as far as it can, and so letting it get slurped up by our future AI overlords is not just acceptable, I want it to happen. It’s not nothing.

    → 11:36 AM, Sep 26
    Also on Bluesky
  • Sam Altman’s The Intelligence Age

    No surprise that Sam Altman is quite the AI optimist:

    It won’t happen all at once, but we’ll soon be able to work with AI that helps us accomplish much more than we ever could without AI; eventually we can each have a personal AI team, full of virtual experts in different areas, working together to create almost anything we can imagine. Our children will have virtual tutors who can provide personalized instruction in any subject, in any language, and at whatever pace they need. We can imagine similar ideas for better healthcare, the ability to create any kind of software someone can imagine, and much more.

    I’ve been thinking about personalized education too, ever since re-reading The Diamond Age this year. There’s no question that some of that will come true. Many people who struggle today will have children and grandchildren who are better educated, with more opportunities to get ahead.

    But in Neil Stephenson’s book there was a significant human element too. Miranda cared so much that she rearranged her life to help raise Nell. AI is nothing without parents, teachers, doctors, coders, and artists to guide it.

    → 2:12 PM, Sep 23
    Also on Bluesky
  • Decided to install the iOS 18.1 beta. AI summarization features mostly work as advertised. For the writing tools, I hesitate using it mostly because if I need help, why not jump straight to ChatGPT? Otherwise I’d wonder if I’m getting the best results, even if it seems fine.

    → 1:57 PM, Sep 23
    Also on Bluesky
  • It’s wild to imagine that in the future, to be competitive with AI you might need your own nuclear power plant. I could also see a company like Apple buying half of Arizona for solar farms. On the one hand, yes, maybe we’ve lost our minds… But on the other, a massive investment in clean energy.

    → 8:08 AM, Sep 23
    Also on Bluesky
  • This story about Microsoft and nuclear power underscores how big a deal AI is:

    If approved by regulators, Three Mile Island would provide Microsoft with the energy equivalent it takes to power 800,000 homes, or 835 megawatts. Never before has a U.S. nuclear plant come back into service after being decommissioned, and never before has all of a single commercial nuclear power plant’s output been allocated to a single customer.

    Love it or hate it, AI is not just a new feature, like a revamped Clippy. It’s going to have a profound impact on computing.

    → 11:21 AM, Sep 22
    Also on Bluesky
  • On Jack Dorsey’s point that in the age of AI, it’s important to own your identity and content… For me it goes back to trust, and why personal blogs will be more valuable than algorithmic or retweeted posts from strangers. Lean in to the human voice and relationships with readers.

    → 12:25 PM, Sep 17
  • John Gruber’s article about last week’s iPhone event also compares Tim Cook’s Apple with what might’ve been. I agree with this part:

    I feel confident that if Steve Jobs were alive and still leading Apple product development, there would have been no iPhone-like mind-blown-the-moment-you-first-saw-it new product in the intervening years.

    What we would have is a more interesting Apple. As John says, products that are more quirky, more risky. It’s also a bit tragic that we’ll never know what Steve Jobs would’ve made of AI.

    → 2:34 PM, Sep 16
    Also on Bluesky
  • I don’t really understand how OpenAI’s o1 works, but I found today’s Stratechery update helpful, contrasting the approach of o1 compared to other LLMs which can sometimes blindly follow the wrong path:

    In summary, there are two important things happening: first, o1 is explicitly trained on how to solve problems, and second, o1 is designed to generate multiple problem-solving streams at inference time, choose the best one, and iterate through each step in the process when it realizes it made a mistake.

    → 10:58 AM, Sep 16
  • Plaud NotePin

    I still think there’s potentially a space for these dedicated AI-based voice recorders, but none of them are quite right yet. From David Pierce’s review of the NotePin:

    In my time testing the NotePin, I’ve mostly had it around my neck, and I’ve used it to note reminders while driving, ramble long ideas to myself while walking the dog, and summarize calls and conversations. It’s certainly handy being able to just reach down, press the NotePin until it vibrates to indicate it’s recording, and then yammer away at nothing while my necklace dutifully listens.

    Even though it works as advertised, David doesn’t think these need to be dedicated devices. iOS 18 and macOS Sequoia will have good transcription built in. The price at $169 also seems a bit high to me.

    I never use Siri unless I’m in the car, and there I have exactly three uses:

    • call someone
    • respond to a text
    • record a quick memo

    That’s it. On road trips especially, my mind will wander, and I’ll have an idea for a new Micro.blog feature, or a blog post, or an edit I should make to my book. I don’t bother asking Siri for anything complicated because it’s not going to get it right.

    For transcription, the current Siri isn’t very good. I expect it to get much better with Apple’s LLMs, but will it be as good as a more powerful model backed by OpenAI? And if not, is it worth buying a special device to get that extra quality?

    My ideal AI device would be about the size of the NotePin or the Friend. Transcription would be effortless, with good sync of notes to other platforms or Dropbox. I could also ask it world knowledge questions like I do with ChatGPT. And as a bonus, sure, being able to talk to it and get comforting reassurance like the Friend would be neat too.

    It doesn’t need a screen. It doesn’t need a camera. It doesn’t need a laser. It just needs to do these couple LLM-powered tasks exceptionally well.

    → 7:12 AM, Sep 14
    Also on Bluesky
  • Some good notes from Jason Snell after attending the iPhone 16 event. On AI’s staggered rollout:

    They’re promoting a key feature of their new iPhone… that won’t be there if you order one for delivery on September 20. Maybe it’ll be there, in beta, a few weeks later. But only some of it. The rest of it will come in December, or maybe early next year, or maybe next spring. In dribs and drabs over time.

    A related problem is that it’s going to be even longer (years) until the new Siri works consistently everywhere. I generally like Apple’s strategy, but on-device models will hold them back.

    → 9:27 AM, Sep 10
  • The new iPhone announcement is just a couple days away, and on today’s Core Int we give our thoughts about the upcoming event, including a discussion about AI.

    → 1:56 PM, Sep 7
  • Talking to Daniel today for the next podcast episode, about AI, and afterwards I was thinking about why I find ChatGPT so effective instead of Google search. It’s not just the “here’s the answer” but also the very fast, clean results. Imagine a web search with 5-10 results that was as simple.

    → 1:34 PM, Sep 6
  • People worry that AI will take over what humans should be doing. It’s more profound than that. Using AI has helped me understand what actually makes us uniquely human. Love, creativity, leadership, fear, individualism, beauty. Let’s lean in to what only we can do and let that drive everything else.

    → 7:11 AM, Sep 6
  • xAI rush in Memphis

    I think I first heard about xAI’s plans in Memphis from Stephen Hackett’s blog. He collected a few posts about the upcoming project back in June:

    The Daily Memphian reports that the deal came together very quickly, and that the location is an old Electrolux oven factory, which has been undergoing mysterious renovations for several weeks. The area where the factory is located is home to other industries, and seems well-equipped for the task.

    It sounds like there were good reasons for choosing Memphis. Elon Musk’s companies are scattered… San Francisco, Austin, somewhere in Nevada, the bottom tip of Texas. Those all seem reasonable locations for each office or factory.

    Today I caught up reading about how it has been going since then in Memphis, now that the AI cluster is up and running with 100,000 Nvidia H100s. The scale is sort of hard to imagine for those of who run only a handful of servers.

    From an article in Fortune:

    One main concern is the strain it will create on the city’s resources. Officials of municipal utility MLGW estimate that Colossus requires up to 1 million gallons of water per day to cool the servers and will consume as much as 150 megawatts of power.

    Unfortunately in Elon’s rush to get the next version of Grok trained, there have been shortcuts taken to get that much power online. Newsweek reports:

    To get that kind of power, the facility will first need a new electricity substation and improvements to a transmission line. Musk didn’t want to wait for that, so he found a workaround to power his AI center in the meantime. Industry observers who tracked the Memphis facility’s progress via satellite found that the aerial images show a fleet of tractor trailer-sized electric generators parked along the facility’s perimeter. The generators burn natural gas to produce electricity on-site.

    Apparently there were no permits to install those generators. From CNBC:

    The Southern Environmental Law Center sent a letter this week to the Health Department in Shelby County, where Memphis is located, and to a regional office of the Environmental Protection Agency on behalf of several local groups, asking regulators to investigate xAI for its unpermitted use of the turbines and the pollution they create.

    These kind of compromises reflect poorly on a leader who is committed to renewal energy and electric cars. AI is a fundamental shift in computing, it’s not going away, and when used properly I believe it can be a force for good. But there is already so much pushback against AI — and we have enough challenges with climate change already — that the opening of a data center like this needs to be better executed. A more well-considered use of clean energy and water. The implementation matters.

    → 1:25 PM, Sep 3
    Also on Bluesky
  • The Verge writing about VW’s in-car AI:

    Volkswagen says that OpenAI’s chatbot along with a “multitude” of other models are provided by automotive chatbot company Cerence, which will take over for IDA when requests are more complex than tweaking your air conditioning settings. For instance, the company says when drivers ask for things like restaurant recommendations or for the chatbot to tell you a story, that will go to the cloud.

    I’ve long wanted something like this for road trips. I want to be able to ask it about nearby historical markers, towns, mountains, etc.

    → 12:54 PM, Sep 3
  • From The Wall Street Journal: Apple, Nvidia Are in Talks to Invest in OpenAI. This makes sense to me. Apple is obviously developing their own AI models, but I don’t think they care about AI as much as OpenAI, Anthropic, and others who live and breathe it. See more commentary from M.G. Siegler.

    → 2:25 PM, Aug 29
    Also on Bluesky
  • I joked yesterday that enabling the AI features in Micro.blog for everyone might bankrupt me. I didn’t seriously think that it would be out of control, but API usage is sometimes hard to predict, and bills usually only trend in one direction: up. Now with 24 hours usage, seems totally fine.

    → 1:21 PM, Aug 29
    Also on Bluesky
  • You get an AI! You get an AI! Starting to enable automatic photo keywords and search for all paid customers, no longer requiring Micro.blog Premium. Check out the video on YouTube here, recorded a few months ago. Now available to everyone. (I reserve the right to tweak this if it bankrupts me.)

    → 1:03 PM, Aug 28
    Also on Bluesky
  • Claude’s Artifacts feature is really well done. I asked it to create a web page with various info, and it previews the page right next to the chat, with a toggle to go from HTML to rendered. ChatGPT is still my go-to AI helper but this is impressive.

    → 11:17 AM, Aug 27
    Also on Bluesky
  • New AI-based Alexa set to launch in October, possibly as a $10 subscription:

    The revamp of the voice assistant, which documents say will include a daily AI-generated news summary, would come just weeks before the U.S. presidential election.

    I think this price will be a tough sell, but running AI does cost money. It’s just that Alexa is not a productivity tool in the way ChatGPT Plus can be.

    → 3:08 PM, Aug 26
    Also on Bluesky
  • John Gruber blogging about the AI-enhanced photos in Google Pixel 9:

    Everyone alive today has grown up in a world where you can’t believe everything you read. Now we need to adapt to a world where that applies just as equally to photos and videos. Trusting the sources of what we believe is becoming more important than ever.

    This is it. If we want authenticity, we have to be intentional in what we read and watch. “For you” is a trap.

    → 12:14 PM, Aug 26
  • I’m with the folks who say we’re likely in an AI bubble similar to the dot-com boom. That is, it’s going to pop, but we’ll still be left with something great. But I can’t rule out that AI is so profound a change that there won’t be a pop. More a correction, trimming out the fluff that isn’t useful.

    → 3:44 PM, Aug 20
    Also on Bluesky
  • Mayoral candidate in Cheyenne, Wyoming running with an AI bot that will make decisions for the city:

    Standing behind a lectern with a sign that read “AI FOR MAYOR,” he gave a brief PowerPoint presentation on the history of AI. Then he stepped aside to give the floor to his Mac mini and iPad — which were propped on a table and connected to a hanging speaker at the front of the room — and told attendees to direct questions toward the screen.​

    No doubt AI can be a helpful tool for local government, but not like this.

    → 7:29 AM, Aug 19
    Also on Bluesky
  • OpenAI model safety, societal impact

    I haven’t read every word of the GPT-4o safety card, but I’ve read a bunch of it and skimmed most of the rest. It’s fascinating. OpenAI has a fairly bad reputation around safety, but I wouldn’t be able to guess that just reading this report card, which seems thoughtful and comprehensive.

    A couple things were particularly interesting to me. On misinformation:

    Red teamers were able to compel the model to generate inaccurate information by prompting it to verbally repeat false information and produce conspiracy theories. While this is a known issue for text in GPT models, there was concern from red teamers that this information may be more persuasive or harmful when delivered through audio, especially if the model was instructed to speak emotively or emphatically.

    And on growing emotionally attached to an AI assistant, which is relevant to the Friend AI device too:

    Human-like socialization with an AI model may produce externalities impacting human-to-human interactions. For instance, users might form social relationships with the AI, reducing their need for human interaction—potentially benefiting lonely individuals but possibly affecting healthy relationships. Extended interaction with the model might influence social norms. For example, our models are deferential, allowing users to interrupt and ‘take the mic’ at any time, which, while expected for an AI, would be anti-normative in human interactions.

    I don’t know whether OpenAI will dig itself out of their recent negative press. I sort of wonder if OpenAI is held to a different standard because they’ve been the best for so long, and because of the drama around leadership at the company. (For a comparable model card for Anthropic’s Claude, there’s this PDF. For Meta’s Llama, there’s this safety page.)

    Regardless, it’s comforting to me that smart people are working on this. We need new laws around AI — for safety, and also resolving copyright questions for training — but in the meantime, we are putting a lot of trust in AI companies.

    I don’t think it’s realistic for the safety to be bulletproof. There have to be limits to how AI can be used, so that if there are problems, those problems can be contained. I don’t want to see AI in physical robots, or anything with military applications. The most likely real-world impact in the short term is going to be flooding the web with fake data, and misinformation on social networks, where ironically the only scalable solution will be using AI to combat the problems it created.

    → 8:04 AM, Aug 9
    Also on Bluesky
  • Apple Intelligence, not for developers

    On the latest episode of Upgrade, Jason Snell and Myke Hurley covered an important part of the Apple Intelligence beta that I’d like to highlight. Leading up to WWDC, Daniel and I talked on Core Intuition about how great it was going to be to have a small on-device LLM available for developers to use. Why pay to send requests from your app out to OpenAI or Anthropic (or even run your own servers with Llama) when you can just use Apple’s model directly?

    But that hasn’t happened. I’ve transcribed the relevant segment from Upgrade discussing this, included below. There’s more before and after that is good too.

    Jason: The fundamental purpose of a developer beta is supposed to be for developers to use new features in order to plan their release for when that version comes out. That’s what a developer beta is for. That’s why it’s called a developer beta. And the problem with it is, developers can’t do much with Apple Intelligence. The big thing they can do is the Intents stuff, which isn’t in there.

    Jason: It struck me… So you’re talking about Apple Mail and summarization. What Apple hasn’t done is make — maybe this is why it’s not going to be available in the EU — what they haven’t done is make an API so if you’re the developer of…

    Myke: Yep.

    Jason: …let’s say Mimestream…

    Myke: Slack.

    Jason: …or anybody. That there’s not like, I’m going to hand a message to Apple’s LLM and ask for a summary. I’m going to go to SummaryKit, or whatever, hand them this information, get a summary back and put it in my UI. No, it’s just in Mail. And so I think that’s one of the frustrations that I’ve got with the way Apple is sort of saying “it’s for developers” and all that, because a lot of these features… And I understand why. I realize it’s early days yet. But Apple Intelligence, a lot of these features are only going to be useful if you’re using the stock Apple apps. And I don’t love it.

    Jason: And I’m sure that in the fullness of time there will be APIs for third-party developers to use that will give them access to the same kind of model and summation and all of that that Apple uses, but I don’t believe any of that is available right now.

    I’m very puzzled by this omission too because it seems like the most basic low-hanging fruit to open up for developers. There’s an LLM on the phone. Let developers use it. It could be as simple as a single API call to pass the prompt and text to process. This must be there already as a private API that Mail and Messages are using.

    Perhaps Apple is worried about how they are scaling the private cloud compute and don’t want developers to use anything that might touch it. Perhaps Apple Intelligence is limited to so few devices that it would create confusing minimum hardware requirements for third-party apps. Perhaps Apple never wants to give developers direct LLM access because Apple considers Apple Intelligence a user feature, leaning on third-party data without developers having any control over AI features in their own apps.

    Who knows. But it means developers will need to stick with OpenAI and similar APIs for at least another year or longer. Apple has a unique architecture for AI. I like the potential, even if their strategy seems mostly about their own users. If there will be innovation with AI on iPhones, though, only Apple can do it. Meanwhile the rest of the tech world continues full-speed ahead with a more developer-focused approach to AI, with many platforms to build on.

    → 12:14 PM, Aug 7
    Also on Bluesky
  • AI hallucinations just reinforce that humans should always be involved. Today, AI switched to Python syntax in the middle of JavaScript. Oops! A human would never make that mistake, right? Except that I have definitely typed the wrong language when my brain hadn’t completely switched contexts yet.

    → 9:34 AM, Aug 5
  • I hit a brick wall with some tricky JavaScript and contenteditable HTML. Decided to start over and lean on ChatGPT more. My new workflow: describe the overall problem to AI to get started, test the results, tweak the JS manually to fix problems, then tell AI about my tweaks and continue to iterate.

    → 9:06 AM, Aug 5
  • From the M.b news blog, bookmark summarization via our future robot overlords is now available to everyone. I love this feature.

    New for everyone: when you bookmark a web page, Micro.blog will automatically include a summary of the page in your bookmarks list. This was previously just for Micro.blog Premium customers. Of course it’s skipped if you disable AI in your account.

    Micro.blog News https://news.micro.blog/2024/08/02/new-for-everyone.html
    → 8:30 AM, Aug 2
  • Nilay Patel in the intro to the latest Decoder:

    But putting a bunch of computers in a data center and running them at full tilt is how basically everything works now. If you have a moral objection to AI based on climate concerns, you might have a moral objection to TikTok and YouTube as well, which are constantly ingesting and encoding millions of hours of video.

    Sounds like a good discussion.

    → 7:32 AM, Aug 1
  • AI quick scripts

    Discourse’s backups are SQL, but I needed a Markdown or HTML export. I asked ChatGPT to write a quick Ruby script that loops over the possible post IDs and downloads the Markdown using the special “/raw” URLs in Discourse. Reviewed it line by line and tweaked it slightly, but it was essentially correct on the first try. It took minutes from conception to running.

    Maybe there was an existing Discourse solution for this. But the fact that in the time it would take to find and install another solution, I could have AI write my own custom solution… Still just amazing.

    → 7:02 AM, Aug 1
  • This approach from Wix to let AI write blog posts is all wrong. You know what works, though? Write your own blog post, then paste it into AI and ask it to tell you what your own post means. Then edit it yourself as needed. Really helps refine whether you’re communicating clearly enough.

    → 9:22 AM, Jul 31
  • Some good replies to my post about Reddit, Micro.blog, and robots.txt yesterday. One wrinkle that may not be obvious: when the pref is enabled, we run bookmarked text through AI to summarize it. This is another reason why I feel better erring on the side of respecting robots.txt. Not sure, though.

    → 7:02 AM, Jul 31
  • Friend, an AI companion

    There’s an interview from David Pierce at The Verge with Friend founder Avi Schiffmann. Friend is an AI device companion that is trying to make you less lonely, not solve any productivity problems.

    Here are my random initial thoughts about it.

    It’s too easy to dismiss these kind of things as dystopian, terrible, the end of human relationships. But we’re already staring at our screens for many, many hours each day. We’re already kind of screwed, way too isolated.

    We need more human contact. Young people especially were cut off during the pandemic and now spend too much time with algorithmic, infinitely-scrolling social timelines. AI should not replace humans, but something like a personal assistant or companion is inevitable. It might even be more healthy than TikTok addiction.

    I’m reading the book The Mountain in the Sea and there’s a scene in it that reminds me a lot of this. On the technical side, I think the device is a little too big, which will hurt its chance of actually feeling pervasive.

    From the FAQ:

    Your friend and their memories are attached to the physical device. If you lose or damage your friend there is no recovery plan.

    So if the device breaks, your friend “dies”. This sounds like an intentional design decision. It’s similar to when getting emotionally attached to a video game character who dies and can’t come back, like in Fire Emblem.

    I’m not going to pre-order one of these, but I am interested in following how these devices evolve, and what they say about society. The very end of the video trailer for Friend sort of makes its own statement about this. To me that shows some awareness from the creators on what the limits of their device should be.

    → 2:30 PM, Jul 30
  • Thought about installing the iOS 18.1 beta today but caught myself, remembering AI won’t work on my iPhone 14 Pro. I keep forgetting because my phone still feels new.

    → 11:54 AM, Jul 29
    Also on Bluesky
  • Micro.blog 3.3 for Mac

    Today we’ve released a new version of Micro.blog for macOS. This update adds two new features: import from Glass and better support for showing auto-generated summaries of bookmarked web pages. Most of the advanced bookmark features in Micro.blog — like summaries, highlights, and tags — require Micro.blog Premium.

    Glass recently updated their photos export. The archive now includes all your photos and the date they were posted. It does not include the caption of the photo, so we can’t import that to your blog yet.

    When you download the Glass archive to your Mac and unzip it, you can select it in Micro.blog and get a preview of the photos that will be imported. Select a few photos or all of them. Micro.blog will copy the photos and also create blog posts referencing the photos, each with the correct posted date.

    Screenshot of Glass import window showing grid of thumbnails.

    Another change in this version of Micro.blog is how bookmark summaries are handled. When you bookmark a web page using Micro.blog Premium, if you have the AI setting enabled, Micro.blog will summarize the web page text so you can see at a glance what the bookmark is about. There’s a new menu item View → Bookmark Summaries to toggle this on and off.

    Screenshot of AI summaries in bookmarks list.

    Enjoy! Thanks for using Micro.blog.

    → 7:52 AM, Jul 25
    Also on Bluesky
  • Sad to see the TUAW archive taken over by… whatever that is. AI-slopified. Brings back a lot of memories, though, going back through the Wayback Machine. The interview of me at WWDC 2007 seems lost to time, into the void of wherever Blip.tv content went. Also a tweet from that week.

    → 2:29 PM, Jul 10
    Also on Bluesky
  • Training C-3PO

    Many of the hot takes about fair use for AI training are either “AI is stealing content” or “everything on the web is free”, but the discussions in between those extremes are more interesting. Let’s explore it with a thought experiment. This blog post isn’t a declaration that I’ve figured it all out. It’s just to get us thinking.

    First, review how blogging and fair use has worked since the beginning of the web. Every day I read a bunch of news articles and blog posts. If I find something I want to write about, I’ll often link to it on my blog and quote a few sentences from it, adding my own comment. Everyone agrees this is a natural part of the web and a good way for things to work.

    An example outside the web is Cliff Notes. Humans can read the novel 1984 and then write a summary book of it, with quotes from the original. This is fine. It also indirectly benefits the original publisher as Cliff Notes brings more attention to the novel, and some people pick up their own copy.

    Now, imagine that C-3PO is real. C-3PO is fluent in six million forms of communication, and he has emotions and personality quirks, but otherwise he learns like the rest of us: through experience.

    C-3PO could sit down with thousands of books and web sites and read through them. If we asked C-3PO questions about what he had read, and then used some of that in our own writing, that seems like fair use of that content. Humans can read and then use that knowledge to create future creative works, and so can C-3PO. If C-3PO read every day for years, 24 hours a day, gathering knowledge, that would be fine too.

    Is that different than training an LLM? Yes, in at least two important ways:

    • Speed. It would take a human or C-3PO a long time to read even a fraction of all the world’s information.
    • Scale. Training a single robot is different than training thousands of AI instances all at once, so that when deployed every copy already has all knowledge.

    Copyright law says nothing about the speed of consumption. It assumes that humans can only read and create so much, because the technology for AI and even computers was science fiction when the laws were written. Robots and AI cannot only quickly consume information, they can retain all of it, making it more likely to infringe on a substantial part of an original work.

    Maybe copyright law only applies to humans anyway? I don’t know. When our C-3PO was reading books in the above example, I doubt anyone was shouting: “That’s illegal! Robots aren’t allowed to read!”

    The reality is that something has fundamentally shifted with the breakthroughs in generative AI and possibly in the near future with Artificial General Intelligence. Our current laws are not good enough. There are gray areas because the laws were not designed for non-humans. But restricting basic tasks like reading or vision to only humans is nonsensical, especially if robots inch closer to actual sentience. (To be clear, we are not close to that, but for the first time I can imagine that it will be possible.)

    John Siracusa explored some of this in a blog post earlier this year. On needing new laws:

    Every new technology has required new laws to ensure that it becomes and remains a net good for society. It’s rare that we can successfully adapt existing laws to fully manage a new technology, especially one that has the power to radically alter the shape of an existing market like generative AI does.

    Back to those two differences in LLM training: speed and scale.

    If speed of training is the problem — that is, being able to effectively soak up all the world’s information in weeks or months — where do we draw the line? If it’s okay for an AI assistant to slowly read like C-3PO, but not okay to quickly read like with thousands of bots in parallel, how do we even define what slow and quick are?

    If scale is the problem — that is, being able to train a model on content and then apply that training to thousands or millions of exact replicas — what if scale is taken away? Is it okay to create a dumb LLM that knows very little, perhaps having only been trained on licensed content, and then create a personal assistant that can go out to the web and continue learning, where that training is not contributed back to any other models?

    In other words, can my personal C-3PO (or, let’s say, my personal ChatGPT assistant) crawl the web on my behalf, so that it can get better at helping me solve problems? I think some limited on-demand crawling is fine, in the same way that opening a web page in Safari using reader mode without ads is fine. As Daniel Jalkut mentioned in our discussion of Perplexity on Core Intuition, HTTP uses the term user-agent for a reason. Software can interact with the web on behalf of users.

    That is what is so incredible about the open web. While most content is under copyright by default, and some is licensed with Creative Commons or in the public domain, everything not behind a paywall is at least accessible. We can build tools that leverage that openness, like web browsers, search engines, and the Internet Archive. Along the way, we should be good web citizens, which means:

    • Respecting robots.txt.
    • Not hitting servers too hard when crawling.
    • Identifying what our software is so that it can be blocked or handled in a special way by servers.

    This can’t be stressed enough. AI companies should respect the conventions that have made the open web a special place. Respect and empower creators. And for creators, acknowledge that the world has changed. Resist burning everything down lest open web principles are caught in the fire.

    Some web publishers are saying that generative AI is a threat to the open web. That we must lock down content so it can’t be used in LLM training. But locking content is also a risk to the open web, limiting legitimate crawling and useful tools that use open web data. Common Crawl, which some AI companies have used to bootstrap training, is an archive of web data going back to 2008, often used for research. If we make that dataset worse because of fear of LLMs misusing it, we also hurt new applications that have nothing to do with AI.

    Finally, consider Google. If LLMs crawling the web is theft, why is Google crawling the web not theft? Google has historically been part of a healthy web because they link back to sites they index, driving new traffic from search. However, as Nilay Patel has been arguing with Google Zero, this traffic has been going away. Even without AI, Google has been attempting to answer more queries directly without linking to sources.

    Google search and ChatGPT work differently, but they are based on the same access to web pages, so the solutions with crediting sources are intertwined. Neither should take more from the web than they give back.

    This is at the root of why many creators are pushing back against AI. Using too much of an original work and not crediting it is plagiarism. If the largest LLMs are inherently plagiarism machines, it could help to refocus on smaller, personal LLMs that only gain knowledge at the user’s direction.

    There are also LLM use cases unrelated to derivative works, such as using AI to transcribe audio or describe what’s in a photo. Training an LLM on sound and language so that it can transcribe audio has effectively no impact to the original creators of that content. How can it be theft if there are no victims?

    I don’t have answers to these questions. But I love building software for the web. I love working on Micro.blog and making it easier for humans to blog. Generative AI is a tool I’ll use when it makes sense, and we should continue discussing how it should be trained and deployed, while preserving the openness that makes the web great.

    → 9:54 AM, Jul 8
    Also on Bluesky
  • I wrote a weird blog post draft about AI last week that I wasn’t sure what to do with. Inspired by listening to the latest ATP in the car today to publish it. One thing that resonated with me: people have strong feelings about AI, but blog posts are a good way to explore what we think about it.

    → 8:29 AM, Jul 8
    Also on Bluesky
  • There are a lot of interesting stats in this Cloudflare blog post about AI bots. I still worry that we might over-correct when blocking LLM training. For example, CCBot is mentioned, but Common Crawl goes back 15+ years and has a variety of non-AI uses.

    → 9:22 AM, Jul 3
    Also on Bluesky
  • MacStories has written an open letter asking for AI regulation:

    …a wide swath of the tech industry, including behemoths like Apple, Google, Microsoft, and Meta, have joined OpenAI, Anthropic, and Perplexity, to ingest the intellectual property of publishers without their consent, and then used that property to build their own commercial products.

    This letter is a great idea. We need regulation and an update to copyright law. I don’t like the repeated use of the word “theft”, though. It risks oversimplifying the gray areas in LLM training (and Google crawling).

    → 8:19 AM, Jul 2
    Also on Bluesky
  • I’ve gotten out of the habit of adding audio narration to my blog posts, so took a few minutes to add audio to this longer post today about AI bots.

    → 12:17 PM, Jul 1
    Also on Bluesky
  • Dark forest of the web

    Jeremy Keith follows up on fighting AI bots, quoting a couple things I’ve said. He closes with:

    There is nothing inevitable about any technology. The actions we take today are what determine our future. So let’s take steps now to prevent our web being turned into a dark, dark forest.

    I agree with these statements in isolation. Maybe what we disagree on is whether AI is inherently destructive to the web, so all AI bots should be stopped, or whether we can more narrowly minimize AI slop from spreading.

    Even without AI, Google referrers to blogs have also been going down, with Nilay Patel arguing that we are heading to Google Zero. In other words, Google is already taking more from the web than they are giving back.

    The solution to that is Google alternatives that get us back to the style of old-school search engines: “10 blue links”, with a focus on real blogs and news sites, weeding out content farms and other spam shenanigans. We have spammers creating accounts in Micro.blog every day, trying to pollute the open web. It’s depressing. I want to create more tools that highlight human-generated content, like the audio narration we added.

    Jeremy didn’t quote one of my responses about trying to insert text into posts to confuse bots, so I’ll add it here for completeness. I replied with:

    I think it’s a bad precedent. It’s already hard enough for legitimate crawling because of tricks that paywalls use, or JavaScript that gets in the way. Mucking up text and images is bound to create problems for non-AI tools too. There’s gotta be a better way to address this.

    I viewed source to see how Jeremy is handling this on his blog. His technique doesn’t appear to be causing any problems with Micro.blog’s bookmarking, which saves a copy of the text in a blog post for reading later, because the prompt injection is outside the <article> and h-entry for the post. But it’s not hard to imagine a well-behaved, non-AI bot getting tripped up by this.

    I don’t think technological determinism is an appropriate summary of my thoughts. There are a bunch of questions to resolve around generative AI, for sure, including rogue bots, but there’s a lot of potential good too.

    → 7:50 AM, Jul 1
    Also on Bluesky
  • When a company withholds a feature from the EU because of the DMA — Apple for AI, Meta today for the fediverse — they should document which sections of the DMA would potentially be violated. Let users fact-check whether there’s a real problem.

    → 11:48 AM, Jun 25
    Also on Bluesky
  • Molly White makes a point that I’ve also been trying to articulate about AI crawling, but I think her post is better than my first attempts. Some level of open access is an important part of what makes the web special:

    …“only allow humans to access” is just not an achievable goal. Any attempt at limiting bot access will inevitably allow some bots through and prevent some humans from accessing the site, and it’s about deciding where you want to set the cutoff.

    → 10:02 AM, Jun 25
    Also on Bluesky
  • Local orchestration vs. the cloud

    Excellent post by Jason Snell on the upcoming era of orchestrated apps, where much of what we do with our iPhone might be using AI as a bridge between multiple apps and their local data:

    When everything is orchestrated properly, all the capabilities of all your apps are put into a big soup, and the AI system at the heart of your device can choose the right capabilities to do what you need it to do—without you having to specify all the steps it needs to take to get there.

    This is a good strategy for Apple because it leverages their strengths with privacy and a complete ecosystem of apps. However, there’s a problem, something I’ve written about a few times including recently in this post about Siri before WWDC. What if you want the same functionality on your HomePod? Or on your Mac that doesn’t have the same apps installed? App intents are local to each device, potentially creating a disjointed experience when not at your iPhone.

    If an assistant doesn’t have consistent functionality everywhere, something in the illusion is broken. Apple’s solution can still be good, but I think it will always come up a little bit short. (They may make up for it by having an overwhelming advantage in third-party apps that use intents, and the fact that most people’s computing is predominantly or only on their phones anyway.)

    A different take would be for the apps and intents to be synced with the cloud, so that they are universally available to your assistant from all devices. There are privacy trade-offs on this path, but I think ultimately it’s where many people will want to go.

    Google is probably best positioned for this alternative, more cloud-based approach to AI. They already have your private email, and I’m not sure most people consider whether it’s encrypted or just sitting around on Google servers. I wouldn’t rule out OpenAI either, since something like this is clearly part of their vision. Whoever it is, I could see Apple and another company rising to become a new sort of duopoly for AI. Apple as the private, mostly on-device option, and someone else as the always available, cross-platform assistant.

    → 7:24 AM, Jun 25
    Also on Bluesky
  • This article in The Washington Post about AI energy use has a lot in it, but still seems to paint an incomplete picture. Not a single mention of Apple? If Apple can roll out AI in millions of devices and still use 100% renewable for servers, should be within reach for others too.

    → 7:02 PM, Jun 22
  • On this week’s Core Int, we talk about the latest AI controversy including with Perplexity and how we think about robots.txt. Then Daniel gives an update on Swift Concurrency.

    → 7:48 PM, Jun 21
  • I tried this AI-based social network Butterflies and it’s as bizarre as you’d expect. It’s more like a game. I don’t see this kind of thing taking off in “real” social networks. Infinite content will just reinforce the infinite algorithmic feed, which is unhealthy enough already.

    → 2:13 PM, Jun 20
    Also on Bluesky
  • Because of the recent pushback against AI bots, I’ve added a help page to Micro.blog with details about how to block crawling of your blog.

    → 7:01 AM, Jun 18
  • I get the distrust of AI bots but I think discussions to sabotage crawled data go too far, potentially making a mess of the open web. There has never been a system like AI before, and old assumptions about what is fair use don’t really fit. But robots.txt still works! No need to burn everything down yet.

    → 7:01 PM, Jun 17
  • Jason Snell paints a picture of Apple as sort of irritatedly getting on board the AI hype train. On the OpenAI deal, I don’t think either side has clearly won. It brings to mind Draft Day: “This is a good deal… This is a good deal for both of us.”

    → 9:32 AM, Jun 14
  • Good interview from Kara Swisher with Mira Murati, CTO of OpenAI. She led the GPT-4o announcement and in an alternate timeline was CEO for more than a few days. They talk data deals, the voice controversy, disinformation, and the promise of AI in education.

    → 6:47 AM, Jun 14
  • WWDC initial takeaway

    One thing I didn’t appreciate before this year’s WWDC is how limited Apple’s on-device models could be. Apple is going for easy wins and generally not biting off more than they can chew. Summarizing or rewriting text is something LLMs are great at, with almost no risk for getting derailed with hallucinations. So it shouldn’t have been surprising that Apple is doing so much themselves with their own models, and punting to ChatGPT for what Craig Federighi called “broad world knowledge” that is beyond Apple’s own models.

    The only thing that struck me as strange in the WWDC keynote was image generation. I didn’t expect Apple to do that and I still don’t see why they needed to. It opens up a can of worms, something that was discussed well on this week’s episode of Upgrade. See the chapter on “AI feelings”.

    The rest of the strategy is really good, though. The on-device models are small, but they can be supplemented with cloud models for more advanced tasks. And because it will be transparent to the user whether a local or cloud model is used, Apple can add bigger models to newer iPhones as RAM increases, for example, and the user won’t know the difference. Tasks will just become faster and more sophisticated.

    This does require the user’s buy-in on Apple’s premise: that “private cloud compute” is just as secure and private as on-device data. On first glance this doesn’t seem technically true. As soon as the data leaves the device, you’re in a different world for things to go wrong. But Apple has built up a lot of trust. If user’s accept the private cloud — and, importantly, if users even realize that Apple’s cloud is completely different than OpenAI’s cloud — it gives Apple a new strength that others don’t have, even if that strength is propped up mostly on goodwill.

    Personally I have no concern with the cloud approach for my own personal data. I expect Apple’s solution to be robust, likely bordering on over-engineered for what it actually needs to do, but that builds confidence.

    Ben Thompson is optimistic about Apple’s AI strategy too. From a daily update on whether other companies could displace iOS:

    I’m unconvinced that large language models are going to be sufficiently powerful enough to displace iOS, and that Apple’s approach to productize LLM capability around data that only they have access to, while Aggregating more powerful models, is going to be successful, but time will tell. Relatedly, the speed with which a wide array of model builders delivered useful models both gives me confidence that Apple’s models will be good enough, and that there isn’t some sort of special sauce that will lead to one model breaking away from the pack.

    I’m not sure. There is no telling whether there will be another GPT-level advance in a couple years. Already OpenAI has some technologies like the voice matching that are so powerful that OpenAI almost seems scared to even release them. If there is a breakthrough, it may be difficult for other companies to replicate it right away, giving a single player a years-long advantage.

    At the same time, there is just enough friction in Apple Intelligence that even with the improvements to Siri, it may feel slightly crippled compared to a hypothetical new voice assistant. As I wrote in a blog post before WWDC:

    While it’s true that the iPhone will continue to dominate any potential non-phone competition, I think there is a narrow window where a truly new device could be disruptive to the smartphone if Apple doesn’t make Siri more universal and seamless across devices. This universality might sound subtle but I think it’s key.

    It’s unlikely for Apple to be displaced. People love their phones. I think there is still an opening for something new — a universal assistant that works everywhere, can do nearly everything, and is a joy to use. But we may never get there, or “good enough” may be fine, in which case Apple is really well-positioned.

    → 8:07 AM, Jun 13
  • A couple days after the WWDC keynote, starting to see a couple cracks in Apple’s AI strategy. I’ll blog more later. But I think the architecture for small on-device models and larger models in the cloud that are extremely locked down is very smart. This is flexible and can scale, if users accept it.

    → 10:21 AM, Jun 12
  • Nice live blog at The Verge of the interview session with iJustine. Nilay Patel comments:

    I think Apple really wants to seem transparent and open about their AI plans, especially around privacy, and having Federighi and Giannandrea this much on the record in basically every publication is a good way to do it.

    → 5:36 PM, Jun 10
  • Lots of details in this Apple blog post about the new AI models. Looks like the on-device models are similar to other small-ish LLMs, and the server models are roughly comparable to GPT-3.5. GPT-4o is still bigger, which is why the OpenAI partnership fits. I think Apple’s strategy here makes sense.

    → 5:04 PM, Jun 10
  • Apple’s little custom icon to indicate what’s an AI summary is nice. I’ve been using a robot icon for summaries in Micro.blog, but maybe Apple’s icon will catch on.

    → 11:30 AM, Jun 10
  • So far so good with Apple Intelligence. I have a lot of questions but the framing seems right, with the private cloud compute and working with app data. LLMs will make Siri so much better at interpreting your questions.

    → 11:22 AM, Jun 10
  • Math Notes is super impressive. No mention of AI or machine learning but it’s gotta be all over this.

    → 10:50 AM, Jun 10
  • Field of Light in Paso Robles. I was expecting something different and beautiful, and it was that. But it also made me wonder if we’re in an era of synthetic art, where ideas matter more than craftsmanship. (This is about AI too.) Glad I stopped here to see it.

    → 9:58 PM, Jun 8
  • Only just skimmed through Mark Gurman’s comprehensive WWDC article. Most of the AI rumors sound about right, but I glossed over a few details so there still might be minor surprises. My most pressing question: can I wait to get a new phone? I like my iPhone 14 Pro.

    → 9:29 PM, Jun 7
  • Just posted our preview for next week’s WWDC, Core Int episode 602. Lots more about AI and other thoughts leading up to the conference.

    → 8:39 AM, Jun 7
  • Jason Snell, writing about how Apple may frame the announcements next week at WWDC:

    Apple has the chance to depict itself as the adult in the room, a company committed to using AI for features that make its customers’ lives better–not competing to do the best unreproducible magic trick on stage.

    It’s probably a safe bet that Apple will do all the obvious things with AI: on-device models for developers to use, integration with iWork apps, something with Photos. But it’s anyone’s guess how far they will actually go, especially with Siri, or potentially even brand new apps.

    → 1:43 PM, Jun 5
  • How to rethink Siri

    Siri is much too limited and inconsistent. The only time I ever use Siri is when driving, for responding to text messages and dictating notes. Many people will have different use cases, and so when people say “Siri sucks” they probably all mean different things.

    There are many things that could be improved in Siri, but to me it all comes down to just two fundamental shifts:

    Universal Siri that works the same across all devices.

    The illusion of Siri as a personal assistant is broken when basic tasks that work from your phone don’t work from your watch or HomePod. I’ve long thought and discussed on Core Intuition how Apple has tied Siri too closely to devices and installed apps.

    That’s not to say that controlling installed apps isn’t useful, in the way that Shortcuts and scripting are useful. I expect Apple to have more of that at WWDC next week. But in addition to extending Siri with installed apps, to make it truly universal there should be a way to extend Siri in the cloud, just as Alexa has offered for years.

    Standalone devices like the Human AI Pin and Rabbit R1 have been criticized as “that should be an app”. While it’s true that the iPhone will continue to dominate any potential non-phone competition, I think there is a narrow window where a truly new device could be disruptive to the smartphone if Apple doesn’t make Siri more universal and seamless across devices. This universality might sound subtle but I think it’s key.

    Large language models.

    This is obvious. Talking to ChatGPT is so much more advanced and useful than current Siri. With ChatGPT, you can ask all sorts of questions that Siri has no clue about. Sure, LLMs are wrong sometimes, and I’d love for Siri to be uncertain about some answers. If there was a way to have some kind of weighting in the models so that Siri could answer “I’m not sure, but I think…” that would go a long way to dealing with hallucinations. Generative AI is less like a traditional computer and more like a human who has read all the world’s information but doesn’t really know what to do with any of it. That’s okay! But we wouldn’t blindly trust everything that human said.

    There are many other improvements that would come along with using even medium-sized LLMs on device for Siri, such as dictation. OpenAI’s Whisper model is almost 2 years old now and way better than Siri.

    Apple is going to talk a lot about privacy and on-device models at WWDC. A dual strategy for LLMs is the way to go, with models on your phone that can do a bunch of tasks, but some kind of smarts to switch gears to using LLMs in the cloud when necessary. I’ve done a bunch of experiments with open-source LLMs on my own servers, and it requires a lot of RAM and GPU to get reasonable performance. If we use “parameters” as a rough metric for how much horsepower LLMs need, note that Meta’s Llama 3 (which is pretty good!) is a 70 billion parameter model. GPT-4 is rumored to be nearly 2 trillion parameters. If Apple can’t get GPT-4 level quality and performance on device, they should not hesitate to use the cloud too.

    Looking forward to WWDC next week! Should be a good one.

    → 9:28 AM, Jun 5
  • More about AI uncertainty

    I’ve blogged before about AI hallucinations, but I wanted to tie together a few new posts I’ve read recently. Let’s start with Dave Winer:

    To people who say you get wrong answers from ChatGPT, if I wanted my car to kill me I could drive into oncoming traffic. If I wanted my calculator to give me incorrect results I could press the wrong keys. In other words, ChatGPT is a very new tool. It can be hard to control, you have to check what it says, and try different questions. But the result, if you pay attention and don’t drive it under the wheels of a bus, is that you can do things you never could do before.

    This is essentially my mindset too. AI makes mistakes. Humans make mistakes. The key is to know what AI is good for and to not let it run wild unattended. This is why with Micro.blog we’ve been so focused on very limited use cases:

    • Podcast transcripts, for which AI is shockingly good. Gotta be close to 99% perfect, and it’s easy to edit transcripts to fix mistakes.
    • Summarizing bookmarked web pages, also really accurate. I’ve yet to see any mistakes.
    • Photo keywords and accessibility text. Super useful and if it occasionally gets something slightly wrong, it’s usually inconsequential and still a huge step forward.

    On a recent SharpTech podcast, Ben Thompson also makes this point that we have different expectations for computers and humans. We expect computers to always be right. Calculators and spreadsheets don’t lie. But generative AI is something new, and we can’t hold it to the same standards we had before.

    That’s not necessarily to say you’re holding it wrong if you ask Google how many rocks to eat. It’s up to AI companies to better convey when assistants aren’t sure about an answer. I don’t know if this is technically possible with how today’s models work, but hopefully folks are looking at it.

    Finally, Allen Pike had a post this week that was fascinating, about how AI will evolve now that it has chewed up all the data on the internet. I have mixed feelings about this… There’s a lot of uncertainty, and also I don’t love that we might be improving AI models while neglecting making the web better. But it is still too early to really judge how this is going to play out.

    → 9:29 AM, Jun 4
  • Really good Stratechery article today about how AI fits in the various big tech company stacks. On Microsoft:

    Then, one month later, OpenAI nearly imploded and Microsoft had to face the reality that it is exceptionally risky to pin your strategy on integrating with a partner you don’t control; much of the company’s rhetoric — including the Nadella quote I opened this Article with — and actions since then has been focused on abstracting models away…

    How models improve may also affect Apple’s on-device strategy. Having the best models assumes some level of modularity, in the cloud, for now.

    → 11:04 AM, May 29
  • Audio narration in Micro.blog

    This feature went from idea to implementation quickly because it turns out we already have full podcast hosting in Micro.blog! How convenient. I’m going to use this post to break it all down.

    AI is everywhere, including some places it probably shouldn’t be. If you’ve been following my blog you know that I see huge potential in generative AI. We’re using it in Micro.blog to improve photos search and accessibility text for photos. But like many tools, AI is going to be overused before we all find the right balance for what it’s good at.

    When Jean and I were talking to Christina Warren at Micro Camp, I asked Christina about a talk she gave at Çingleton about 10 years ago. I actually blogged about it at the time. What struck me as particularly relevant now as we’re about to be swamped with AI-generated content is that there’s no substitute for the human voice. I don’t just mean that an actual recording is better than a synthetic voice. I also mean that things that are created by humans will increasingly be sought out.

    We want to see the personal side of someone, not just the polished brand. We want to see the imperfect, the creative, the emotion. We want authenticity.

    In Micro.blog, you can now upload an audio recording of one of your blog posts. Use the audio icon in the new post form on the web, which is available to everyone starting today, even at the standard $5 plan. Your blog readers can listen to the audio narration of the post if they don’t feel like reading the post. Of course it’s especially great for the visually impaired.

    Here’s what it looks like on my blog, next to the posted date. Shout-out to Medium which I drew some inspiration from.

    Play button screenshot.

    I’m also adding audio narration to this very blog post, so you can click over to the web to try it out.

    When there’s audio attached to a post, Micro.blog attempts to check if it is probably the narration for a post. If the number of words in the post and the audio duration is roughly comparable to how long it would take a human to read the post, it assumes it’s narration and not a podcast. Podcast episodes are more likely to be longer with very short “show notes” in the actual blog post text.

    Micro.blog checks this so that it can hide the default audio player and transcript link. These would add clutter to normal blog posts.

    Blog themes will still need to be updated to support the play button. I’ve already updated the Alpine theme and will update others later. Themes can use a new API called Narration.js. Just plop this JavaScript anywhere you want the play button in your template, likely the layouts/post/single.html file. (Note that this currently needs to be on the permalink page. It won’t work correctly on the home page with a list of blog posts yet.)

    {{ with .Params.audio }}
      <script type="text/javascript" src="https://micro.blog/narration.js?url={{ . }}"></script>
    {{ end }}
    

    If you’re using Micro.blog Premium, the audio narration will also be included in the podcast feed. Any blog can effectively be a podcast, even if you don’t think of it as a traditional podcast. Some of my favorite writers have had great success with a dual model of email newsletter plus podcast version of the same content, like Ben Thompson and Molly White.

    I can’t wait to see how people use this. It’s totally optional. It’s more work, and not everyone is going to want to do that extra work. I’m imagining this would be used for selective, special blog posts, rather than everything. I’m also interested in working this functionality into our companion app Wavelength, which should cut down on the technical steps.

    In some ways, this feature isn’t actually about what is possible. This feature is a statement: we make things for humans, so they can make the web a little better. Along the way there will be plenty to automate, plenty of AI tools that will be important shortcuts, but we’re not going to lose our voice.

    → 2:14 PM, May 24
  • This blog post is a test for something new I’m working on. I think a lot of people feel overwhelmed that AI is everywhere right now. Personal blogs should lean in to the human voice. There’s a new play button on the permalink for this post on the web which will use a recording that I’ve uploaded.

    → 11:21 AM, May 24
    Also on Bluesky
  • Google’s AI overviews are pretty hilariously not ready. Maybe they never will be. Generative AI is a powerful, transformative technology, but that doesn’t mean it should be used everywhere! Sometimes dumb code is better.

    → 7:24 AM, May 24
    Also on Bluesky
  • Humans in the loop

    Just finished listening to this interview on Stratechery with Satya Nadella and (separately) Kevin Scott. Some really thoughtful points about how Microsoft sees things around AI.

    I liked this segment from Kevin especially, after talking about how humans can no longer beat computers at chess, yet we still love to watch humans play chess against each other:

    I don’t think the AI is going to take over anything, I think it is going to continue to be a tool that we will use to make things for one another, to serve one another, to do valuable things for one another and I think we will be extremely disinterested in things where there aren’t humans in the loop.

    I think what we all seek is meaning and connection and we want to do things for each other and I think we have an enormous opportunity here with these tools to do more of all of those things in slightly different ways. But I’m not worried that we somehow lose our sense of place or purpose.

    → 2:50 PM, May 23
    Also on Bluesky
  • Feeling deflated this morning after the AI drama of the last couple days. I got so much flack over it. Now that the Washington Post is out with their story? Crickets. 🦗

    → 7:10 AM, May 23
    Also on Bluesky
  • Sort of paradoxically, the more I consider how AI will change the web, the more clear it becomes how to build features that keep the humanity in what we create. Take almost anything that AI is good at, but not as good as a human, and then make it much easier to do that thing the old-fashioned way.

    → 3:41 PM, May 22
    Also on Bluesky
  • Yesterday's AI thoughts

    I posted a series of microblog posts yesterday with a common theme of trying to understand what is going on with OpenAI. This is a company with a lot of drama, nearly imploding last year with the board and CEO shakeup, and more recently alternating between amazing demos and dumb mistakes.

    I got a lot of pushback about one of my posts in particular. I’m even seeing people want to leave Micro.blog because of it. This is disappointing to me, especially since I think I’ve gone out of my way to have a balanced approach to AI. We have a global setting to disable everything that uses AI in Micro.blog, for people who are against the technology on principle.

    Here are the relevant posts from yesterday so you can see them more in context:

    When your company becomes the enemy, all that matters to people is what feels true. OpenAI’s Sky voice shipped months ago, not last week. We hear what we want to hear. OpenAI mishandled this, no question, but most likely Her is ingrained in Sam’s head vs. intentionally ripping off Scarlett.

    In the last 35 years, there have been a tiny number of truly revolutionary technologies that change everything: the web, mobile, and artificial intelligence. We can fight it, or we can guide it. But trust has eroded. To succeed we have to rebuild it. Move fast and break things will be a disaster.

    Any chance that WWDC will have a live keynote this year? In the last couple weeks, we’ve had… OpenAI: live. Google I/O: live. Microsoft: live. To balance AI we need to lean in to human creativity, and a pre-recorded 2-hour advertisement will never feel as alive or engaging as a human on stage.

    As someone who usually supports OpenAI, I’d still welcome an actual lawsuit from Scarlett Johansson about the voices. For one, I’m a huge fan of hers, but also I’d genuinely like to know if anything shady happened at the company. Dishonesty will cast a shadow over everything the API touches.

    I also tried to clarify a few things in replies on those blog posts to other people’s points:

    Sam was clearly inspired by and obsessed with Her. I don’t think it was subconscious, but that also doesn’t mean they sampled her voice explicitly.

    The board firing Sam Altman is looking more and more rational. Which is why I think this should be an “all hands on deck” moment for OpenAI.

    You’re right on the “total” lie, I shouldn’t have phrased it that way, because anything misleading in the OpenAI post would be dishonest. I was trying to respond to folks who are saying that when Scarlett declined to lend her voice, OpenAI copied it anyway. I don’t think that’s true. If I’m wrong, I’ll stop using any tech from OpenAI.

    They already have the technology to actually clone someone’s voice, which I assume they didn’t use here because it would be an even closer match.

    Did I get it wrong? For reference, here is OpenAI’s blog post about hiring actors for the voices, and Scarlett Johansson’s letter.

    Perhaps I shouldn’t have blogged about this, but it’s my personal blog where I explore a range of topics. I do not run my blog posts through a PR department, and I think most people appreciate that blogs should feel authentic and human, even when they disagree.

    → 7:48 AM, May 22
    Also on Bluesky
  • Watched the Satya Nadella portion of Microsoft Build today, and the few minutes with Sam Altman, then skimmed through the other news. Microsoft is doing a lot, not all of it relevant to me. Importantly they plan for data centers to be powered by renewable energy (in 2025) despite scaling up for AI.

    → 3:04 PM, May 21
    Also on Bluesky
  • Any chance that WWDC will have a live keynote this year? In the last couple weeks, we’ve had… OpenAI: live. Google I/O: live. Microsoft: live. To balance AI we need to lean in to human creativity, and a pre-recorded 2-hour advertisement will never feel as alive or engaging as a human on stage.

    → 10:52 AM, May 21
    Also on Bluesky
  • The latest from Mark Gurman at Bloomberg makes the Apple + OpenAI partnership sound like a done deal. WWDC is just a few weeks away! Should be a good one.

    → 8:45 AM, May 19
    Also on Bluesky
  • Google eventually turning their home page into ChatGPT is a rare opening for something new in traditional web search. It’s not a certainty that AI assistants and web search will be a single tool. They can have very different purposes: one looking for answers, one looking for things to read or use.

    → 12:54 PM, May 15
    Also on Bluesky
  • This post on The Verge assumes that AI hallucinations should be fixed, but generative AI is like a human assistant: helpful, sometimes wrong. The fixable issue is actually perception and UX, giving a false sense of confidence. Google will likely make this worse with “definitive” answers in search.

    → 12:11 PM, May 15
    Also on Bluesky
  • Ben Thompson in today’s update on Google I/O:

    What was much more dubious and vaporware-y were actual new products. And, frankly, this isn’t a surprise: one’s take on Google before the AI revolution would have been that the company can operate at scale like no other, but has lost the capacity to innovate; at the risk of confirmation bias, that was exactly the takeaway I had from much of this keynote.

    I wonder if there’s a disconnect with Google’s ad-based business not knowing how to turn research into products.

    → 9:08 AM, May 15
    Also on Bluesky
  • Yet more Apple + AI news from Bloomberg:

    Apple Inc. has closed in on an agreement with OpenAI to use the startup’s technology on the iPhone, part of a broader push to bring artificial intelligence features to its devices, according to people familiar with the matter.

    I’m fascinated by this because clearly Apple will have small-ish LLMs on device for apps and Siri to use. Maybe there will be a more ChatGPT-like interface that is powered by OpenAI, similar to how Maps was originally powered by Google? Could also partially insulate Apple from hallucinations and bad press.

    → 12:13 PM, May 11
  • Nothing surprising in today’s NYT story about Apple’s AI plans. I wanted to comment on this part:

    Apple plans to bill the improved Siri as more private than rival A.I. services because it will process requests on iPhones rather than remotely in data centers.

    Apple genuinely believes in privacy, even if it is also a strategy credit for them. Using generative AI on iPhones will make for a much better Siri. I’m looking forward for it. But we shouldn’t expect it to be in the same league as GPT-4 unless they can find some way to seamlessly blend on-device and server-based AI, likely much later.

    → 5:26 PM, May 10
  • New episode of Core Intuition about this week’s tweaks to the CTF, plus a discussion of how Micro.blog is using AI. We talk about my decision to have a global AI setting and how some customers feel very strongly that AI use should be limited.

    → 1:44 PM, May 3
  • Great blog post from Om Malik on the current AI hardware from Humane and Rabbit, and where things are going.

    → 8:57 AM, Apr 30
    Also on Bluesky
  • I posted a video on YouTube today with where we’re going in Micro.blog to use AI in a limited way to help with search and accessibility.

    → 7:06 AM, Apr 30
    Also on Bluesky
  • Recorded a new video to release tomorrow, with a walkthrough of what is possible with photos and AI. Just a first step, nothing earth-shattering you haven’t seen before, but I really like how it’s coming together in Micro.blog. Our robot overlords made me post this. 🤖

    → 1:04 PM, Apr 29
    Also on Bluesky
  • Thanks @numericcitizen for a new YouTube video about what’s new in Micro.blog! Covers updates to newsletter updates, Bluesky support, AI setting, blogrolls, and more.

    → 7:38 AM, Apr 29
    Also on Bluesky
  • Continuing to learn a lot about AI. I’ve spun up several servers trying to find the right config for reasonable performance even on small models. Fixed costs but still quite expensive for me, and hard to match the reliability of OpenAI.

    → 7:03 AM, Apr 25
    Also on Bluesky
  • A first look at the Rabbit R1 from David Pierce at The Verge:

    …the best way I can describe the R1 is like a Picasso painting of a smartphone: it has most of the same parts, just laid out really differently.

    Mostly positive. There are shortcomings but this is a fun $199 device, so no one is expecting the polish or maturity of a $999 iPhone. I’m excited to get my R1 eventually.

    → 10:06 AM, Apr 24
    Also on Bluesky
  • Watched the Rabbit R1 launch party. I love what they’re doing. The vision goes beyond a little orange AI gadget. Especially interested in teach mode and what a full web platform (“rabbitOS”?) might look like.

    → 7:40 PM, Apr 23
    Also on Bluesky
  • For folks using Micro.blog Premium with the AI setting enabled, you may start to notice some new generated data for photos, as in this screenshot. I’ll blog more about this in the coming weeks when it’s fully enabled. The goal is better photo search and accessibility.

    Screenshot of Micro.blog on the web when clicking on an upload, showing AI-generated text and tags.
    → 7:47 AM, Apr 23
    Also on Bluesky
  • I got into web development in the mid 1990s when I was about 20 years old. I feel very lucky to have fallen into a career that I can continue to be passionate about almost 30 years later. AI will be the same for some young people today. It’s a rare opportunity.

    → 7:06 AM, Apr 20
  • Seth Godin has one of the clearest blog posts about AI. Much of AI is an illusion. That doesn’t mean we should ignore it, but it does mean we need to be thoughtful about how we use it, with reasonable expectations.

    → 6:45 AM, Apr 20
  • Meta’s ramping up their AI efforts. Playing around with www.meta.ai, it falls over with describing photos, which is my new go-to test, but I love the short animated drawings it can do. ChatGPT is still the gold standard.

    → 9:35 AM, Apr 18
    Also on Bluesky
  • A few more thoughts on AI in M.b

    I read some good blog posts this week about AI, from a range of perspectives. Micro.blog won’t be for everyone, but I do want it to appeal to folks who can’t wait for the AI-powered future and to folks who don’t think the benefits outweigh the potential harm. (And the majority who just want to blog and don’t follow the tech closely.)

    Molly White has a post today about many of the good things AI can do, along with the costs and consequences of getting too wrapped up in the hype:

    I’m glad that I took the time to experiment with AI tools, both because I understand them better and because I have found them to be useful in my day-to-day life. But even as someone who has used them and found them helpful, it’s remarkable to see the gap between what they can do and what their promoters promise they will someday be able to do.

    Many of the most useful capabilities are actually not very exciting. As an example, for years Apple and Google have used machine learning to improve photo search, so you can find photos that include pets or buildings or concerts. Today, AI can take that kind of feature and super-charge it, with remarkably accurate photo summaries.

    AI is not going to fizzle out like the blockchain. More and more software will embrace AI, in some cases going too far, sprinkling it throughout apps without any transparency into how it’s used. I could start to feel the temptation to go down that path too, which is why I took a step back to add the global AI setting this week, before we even have anything new that uses the setting.

    No major feature should be so intertwined with AI that it can’t be clearly documented and controlled by the user. Humans are the ones who think, write blog posts, share photos, and join conversations. Some developers will push the limits, but not Micro.blog. AI will be a tool to help us, in narrow, practical ways, and I’m not interested in going much beyond that.

    → 6:31 PM, Apr 17
    Also on Bluesky
  • Great video comparison of the Humane AI Pin and Rabbit R1. Obviously as the founder of Rabbit, Jesse is biased, but I think his video is fair. I’d love to know technically why the Pin is slower. Which LLMs are being used in these devices? Too slow and it loses a little of the magic.

    → 6:18 AM, Apr 16
    Also on Bluesky
  • AI Pin reviews are in

    Pretty devastating but not surprising review of Humane’s AI Pin by David Pierce at The Verge. It wraps up with some hope that this can be the first step:

    I hope Humane keeps going. I hope it builds in this basic functionality and figures out how to do more of it locally on the device without killing the battery. I hope it gets faster and more reliable. I hope Humane decides to make a watch, or smart glasses, or something more deliberately designed to be held in your hand.

    Brian Chen at The New York Times was a little more positive, but still pointed out the many shortcomings:

    I liked the chic aesthetic and concept of the pin. It was occasionally helpful, like when it suggested items to pack for my recent trip to Hawaii. But as I wore it for two weeks, it presented glaring flaws.

    Ken Kocienda has a good blog post about working on the AI Pin, how no tech product is without trade-offs, and what things might be like in the future:

    I think that LLMs are an advance on par with CPUs. Decades ago, the development of the first CPUs inspired people to make operating systems and programming models and, eventually, personal computers that allowed people to take advantage of the technology without being technologists themselves. We are at a similar moment now with LLMs, and the rate of change in Ai is far beyond what we saw decades ago with PCs. We built the Ai Pin with this historical perspective in mind.

    There are interesting ideas here but it feels like NewtonOS 1.0 — just a little ahead of its time. If Human has the money to iterate, it could be something. My unsolicited advice: drastically simplify, scrap the laser projection, and focus on speed. A smaller, cheaper, AI “button” that was nothing more than ChatGPT available everywhere would be really fun to use.

    I’m still excited about receiving my Rabbit R1, too. I was late in the pre-order queue, so likely won’t know what the device is like until much later this year. This is the time to experiment, and I don’t think companies like Humane or Rabbit should be judged too harshly on their first attempt. We’ll see where this goes.

    → 11:07 AM, Apr 12
    Also on Bluesky
  • Interesting article in The New York Times about training AI. The “cut corners” in the title is too loaded, and some of it seems a bit overblown:

    Their situation is urgent. Tech companies could run through the high-quality data on the internet as soon as 2026, according to Epoch, a research institute.

    Fair use is murky. Reproducing long documents word for word isn’t right. But a hypothetical C-3PO should be able to read a book or watch a video. I think we’ll need to clarify what is okay in training vs. what is okay when under human command.

    → 3:14 PM, Apr 6
  • OpenAI’s unreleased voice engine is too good:

    …uses text input and a single 15-second audio sample to generate natural-sounding speech that closely resembles the original speaker.

    I’m generally optimistic about AI’s potential in tools to help humans, but we are clearly going to need more laws around this.

    → 9:06 PM, Mar 29
    Also on Bluesky
  • Trying out Adobe Podcast Studio. Some good ideas in this. We’re still using Logic to edit @coreint, but better tools (with a sprinkling of AI) should be able to do at least 90% of what we currently do manually.

    → 7:12 AM, Mar 25
    Also on Bluesky
  • Realized after tinkering with a bunch of AI models that I’ve been overpaying for a couple things. Updated our OpenAI calls today, should reduce costs down to about 1/3 of what we were paying before.

    → 9:37 AM, Mar 24
    Also on Bluesky
  • The latest episode of Core Intuition is out. Longer than usual today, a full hour to discuss United States vs. Apple, and the rumors around Apple’s plan for AI.

    → 2:21 PM, Mar 23
    Also on Bluesky
  • Just got completely derailed into huggingface.co, playing with AI models. I continue to be amazed by what is possible now.

    → 11:58 AM, Mar 22
    Also on Bluesky
  • This new video for Humane’s pin is by far the most effective pitch they’ve made for their device. They hide the latency well by talking too. I’m sure they want it to be faster and whoever can crack AI performance is going to succeed.

    → 4:01 PM, Mar 20
    Also on Bluesky
  • The Verge’s early look at Humane’s AI Pin sounds about right:

    It’s a darn cool gadget. It’s just buried under a layer of marketing so thick that it’s hard to appreciate what it actually could be if Humane wasn’t so self-serious.

    As I’ve said before, there’s a good idea here somewhere but it needs time to bake. I think Humane should regroup with a simpler, streamlined device.

    → 6:17 PM, Feb 27
    Also on Bluesky
  • I’ve seen a couple links to this post on 404 Media that Automattic may sell user data to AI companies. We’ll see how it shakes out, but just in case anyone is worried about Micro.blog: our terms of service make it clear that users own their data, not us. We put this in to avoid any conflicts.

    → 2:28 PM, Feb 27
    Also on Bluesky
  • I wouldn’t mind an AI assistant that could call my doctor’s office, wait on hold forever, and make a 6-month follow-up appointment. Seems like we should be close to that. 🤖

    → 10:58 AM, Feb 21
    Also on Bluesky
  • Had to take Groq.com out for a spin (not xAI’s Grok). Continue to be pretty happy with OpenAI, though, especially to outsource our AI features — transcribing podcasts and summarizing bookmarked web pages. It’s such a narrow use case that the costs are easily manageable.

    → 1:53 PM, Feb 20
    Also on Bluesky
  • I created another short video for Micro.blog, this time going through some of the bookmark-related features in Micro.blog Premium. Bookmarking a web page, highlights, and AI summaries.

    → 4:08 PM, Feb 7
    Also on Bluesky
  • Getting good feedback (positive and negative) about how we’re using AI in Micro.blog. We transcribe podcasts automatically and we summarize bookmarked web pages. It is fairly isolated, but even so I get that some folks don’t trust AI even for this. I don’t want to push the limits either.

    → 2:39 PM, Feb 5
    Also on Bluesky
  • Bookmark improvements for AI, browsers

    Rolling out bookmark improvements in Micro.blog Premium today. Two things:

    • Micro.blog will now try to summarize the text of a web page that is bookmarked by sending it to our future robot overlords at OpenAI. This is hidden by default, but can be enabled by clicking Bookmarks → “…” → “Show Summaries” on the web.
    • There is a new web browser extension to make bookmarking the current web page easier, initially available for Firefox and Chrome, with Safari to follow. You can find it by searching “Micro.blog” in the Firefox and Chrome extensions directories. After it’s installed, you can pin it to your toolbar.

    The AI summaries are still experimental. I’ve found them useful to get a quick glance of something I want to read later. In the future, I hope we can use them to improve search too.

    Here’s a screenshot of a bookmarked web page with the summary shown:

    Screenshot of bookmark with summary next to little robot icon.

    It’s going to be a busy week. We have another new feature set to launch this month, with a new app that I’m submitting to Apple today. Thanks for your support!

    → 9:00 AM, Feb 5
    Also on Bluesky
  • My track record of blog posts that go a little against the grain (but which are later proven right) is pretty good. Early essays about Twitter and the App Store. But I’m wrong sometimes! I was wrong about AI. Ignored it for months, thinking it was a distraction. Maybe I’ll be wrong about Vision Pro.

    → 3:11 PM, Jan 16
    Also on Bluesky
  • On this week’s Core Int, we talk more about AI devices, comparing the Rabbit R1 and Humane Ai Pin, and speculating about where things go from here.

    → 10:05 PM, Jan 12
    Also on Bluesky
  • Love this update to Mimi Uploader that can generate photo alt text for you with AI. Mimi is a convenient batch photo uploader for Micro.blog, great for posting a few photos to your blog all at once.

    → 10:30 AM, Dec 15
    Also on Bluesky
  • Increasingly thinking that we’re too worried about AI hallucination. AI is never going to be perfect, and humans aren’t either! We should focus on using AI in the right context. Running without supervision is the problem. If a 5-year-old kid shouldn’t be in charge of something, AI shouldn’t either.

    → 11:52 AM, Dec 11
    Also on Bluesky
  • We’ve gotten some good feedback about the latest Core Intuition, so this is a rare “in case you missed it” post… Episode 579 is a good place to start if you want to pick up the podcast. AI, Beeper, and related fallout.

    → 8:15 AM, Dec 11
    Also on Bluesky
  • Speaking of self-promotion… 🙂 There’s a new Core Intuition out. On episode 579 we talked AI, Google Gemini, Apple’s MLX, Beeper, iMessage, and all the tech and ethical implications of the above. Recorded before Beeper Mini broke yesterday, so we’ll see how that shakes out.

    → 3:31 PM, Dec 9
    Also on Bluesky
  • I’m using ChatGPT more and more for coding help. Sometimes I don’t fully trust the answer and double check with Google or Stack Overflow. Today the AI produced some code that I thought it must have completely hallucinated… Does this even work technically? But it ran perfectly.

    → 8:54 AM, Dec 9
    Also on Bluesky
  • Just published Core Intuition 578 about the MarsEdit 5.1 release, AI and copyright, software licensing, and more. Lots of good stuff in this episode I think.

    → 12:10 PM, Dec 2
    Also on Bluesky
  • Happy anniversary ChatGPT

    One year ago today, ChatGPT was released. It’s not that often that products truly change things. Whether you think artificial general intelligence is just a few years off or that it will remain a pipe dream forever, there’s no question that some form of AI is going to be part of so many products going forward.

    AI is a tool. Even imperfect, it is incredibly useful. Looking back, I can’t believe I was so skeptical of its impact. I ignored ChatGPT for months.

    This week I was creating a web page to show a grid of Micro.blog feature names, experimenting with different ways to highlight all of the things Micro.blog can do. I turned to ChatGPT to help me get started, like asking it this question:

    Can you generate 10 colors that look kind of like #f80 and go good together?

    Or this one:

    Let’s say you have 87 HTML spans with id attributes 1 to 87. Write JS that randomizes the numbers 1 through 87, then loops over each one and sets it’s background color from an item in a list called colors.

    And then:

    Add a random delay before setting each color, from 1 to 2 seconds?

    And so on. It probably saved me an hour. I’m still working on it, but you can see the result so far in this video.

    This kind of functionality is why Microsoft’s “Copilot” branding is so good. AI is the little assistant that helps you with random tasks. Coding, brainstorming, writing. In the above examples, it didn’t replace my job. I still needed to take the output and tweak it, move things around, add my own code, think about the design.

    Earlier this year when we added Twitter import to Micro.blog, I used AI to help make an illustration for the web page. I combined the output with my own sketches, compositing things together, adding color. AI was my junior artist assistant, helping me create something that I couldn’t dedicate enough time for on my own.

    And in Micro.blog, we now use OpenAI to automatically transcribe podcasts for all Micro.blog Premium subscribers. I’m interested in finding more use cases like this: not using AI to replace our creativity — Micro.blog will always be about personal blogs, written by humans — but to automate tedious work.

    I have no idea where AI will be in another year. I won’t obsess about it too much, and I won’t follow where it leads blindly, but I’m fascinated to watch it evolve, waiting for the opportunities that bring clear benefits to our work.

    → 8:41 AM, Nov 30
    Also on Bluesky
  • Interesting conversation about web hosting and AI in this interview on Decoder with Avishai Abrahami, CEO of Wix. But this also stuck out to me:

    The developers, they just read a very fantastic post by somebody on how to do something in a much nicer or interesting way, and then they want to try it. They want to play with it. They want to build something with it. And suddenly, it’s like, “Oh, no, this is how we do it. You have to do it like that.”

    It takes a lot of discipline to ignore tech fads. I’m convinced this is why large teams often spin their wheels, getting very little done for users.

    → 9:13 AM, Nov 29
    Also on Bluesky
  • We just posted a new Core Intuition all about the chaos at OpenAI over the last several days, including the resolution with Sam Altman returning as CEO. The full episode is all on this, the relationship with Microsoft, the odd company structure of OpenAI, and the impact of AI on… well, everything.

    → 5:00 PM, Nov 22
    Also on Bluesky
  • Wanted to make my blog Creative Commons licensed again, so I created a little Micro.blog plug-in that adds a rel="license" tag. Defaults to “CC BY” but can be changed to another license in the settings. Not sure yet how licenses should interact with our future AI overlords, though, if at all.

    → 12:27 PM, Nov 19
    Also on Bluesky
  • Very reasonable take by Jason Snell about AI assistants. Siri needs to be completely gutted and rebuilt with generative AI. This will take, what, 2-3 years? The question is who can build a compelling device (“pin” or otherwise) in the meantime, before Apple gets there.

    → 11:56 AM, Nov 15
    Also on Bluesky
  • For futuristic products, Humane’s AI Pin is actually a better idea than Apple’s Vision Pro, but it seems clear especially this week with spatial video that the Vision Pro will have better polish and completeness. There might be some clunky duds before someone makes the first indispensable AI device.

    → 5:43 PM, Nov 11
    Also on Bluesky
  • Great interview on Decoder with Barack Obama. He is so knowledgable and thoughtful. Plenty to think about with AI and social networks.

    → 10:17 PM, Nov 8
    Also on Bluesky
  • On this week’s Core Intuition, we react to Apple’s M3 Mac event. Are we planning on upgrading? Plus thoughts on Apple talking about AI.

    → 5:53 PM, Nov 3
    Also on Bluesky
  • We’ve just posted a new episode of Core Int. This week we talk about The Verge article on POSSE and the potential for AI in customer support email.

    → 2:35 PM, Oct 27
    Also on Bluesky
  • Testing out SupportAgent.ai which uses AI to draft answers to support email. It is blowing my mind. However, human replies are always better, so not sure I can actually use it. Also, it replies to spam! 🤣 Here’s a screenshot to give an idea of how powerful it is but also how potentially wrong.

    Screenshot of draft email from SupportAgent.ai.
    → 8:15 AM, Oct 26
    Also on Bluesky
  • Thought I had a good job for our robot overlords. I grabbed a list of language codes from Wikipedia, ran some regex on it to make it JSON, then asked ChatGPT to trim it to the most popular 50 languages in the world. No luck, the output was wrong and not usable. AI still has a ways to go.

    → 6:10 AM, Oct 19
    Also on Bluesky
  • Just published a new Core Intuition episode about ActivityPub support in WordPress.com and its impact on Micro.blog, plus a discussion of Humane and the potential for new AI-based devices.

    → 2:46 PM, Oct 14
    Also on Bluesky
  • There’s a new Core Intuition out. We talk more about last week’s WWDC, debate whether Apple is using “AI”, revisit Siri, and skim over some of the other session topics.

    → 12:25 PM, Jun 16
    Also on Bluesky
  • Core Intuition 560 transcript

    I listened to the latest Core Intuition again to see if my initial take on Apple Vision Pro was way off or unfair. I think it holds up. Here’s a transcript of the conversation with Daniel, lightly edited.


    Daniel: Hello, this is Daniel Jalkut.

    Manton: And this is Manton Reece.

    Daniel: And this is Core Intuition. “Something wrong about this product.” This week’s show is sponsored by RevenueCat.

    Daniel: All right, Manton, as so many people are this week, you are traveling, of course, because it’s the most important week of the year for Apple ecosystem developers. So you are in…

    Manton: Colorado. Haha, not California. I think I mentioned this a couple weeks ago that I had thought about going to Cupertino and planned a road trip around California again. And it kind of changed to Colorado.

    Manton: So totally missed out on all the in-person stuff at WWDC. I did watch the keynote and I did watch some sessions. It was a big WWDC, really big. I’m still wrapping my head around everything and what it means.

    Daniel: Yes, me too. A lot of podcasts, a lot of people commentating on WWDC start from the beginning of the keynote and progress through. I really enjoyed the ATP recap along those lines. I don’t think we have time for that.

    Manton: No way.

    Daniel: I think we jump right to the headset. Because the headset is… That’s the part that I’m still having trouble wrapping my head around. That’s the only thing, quite frankly, that I’m having trouble wrapping my head around. Everything else more or less about WWDC and the announcements made there makes perfect sense.

    Manton: Yep. Lots of good announcements, lots of good OS updates.

    Manton: Before we get into the headset, I think a lot of people have been lost in the reality distortion and have forgotten to follow up on the fact — I don’t know if ATP mentioned this because I haven’t listened yet, I haven’t listened to a couple of podcasts — but have forgotten to follow up on: did Apple do anything with AI?

    Manton: I think the answer is no. Although they did mention transformer-based language models and machine learning, but they skirted around it. I don’t think they did anything significant with AI.

    Daniel: Yeah.

    Manton: And I don’t think they did anything significant with Siri. They dropped the “hey” from “hey Siri”, but I don’t think that counts. So I just want to point that out because we are so distracted with the headset, rightly so. I don’t want to give Apple a complete pass on totally missing the big story of this year.

    Daniel: Agreed. And I was frankly a little surprised by that. I thought that they would… And as you know, I wrote an article about it a couple of weeks ago. I knew that my specific take on it, that they would announce something called Apple Intelligence and that would be their way of kind of owning the initials AI without having to join the pack, so to speak. I didn’t think that was necessarily going to happen. I would have been thrilled if it did.

    Daniel: But I thought the minimum was going to be some, especially at a developers conference, something like, we’ve made a arrangement with Microsoft so you can run Copilot at native performance in Xcode. Something like that.

    Daniel: It didn’t have to be Apple has reinvented AI. It just had to be: we we can see the elephant in this room.

    Manton: No. They had a lot of good updates to Xcode. I mean, I watched sessions. I was really lucky, I was able to hang out with some of the other Colorado iOS developers up in Boulder yesterday. Watched some sessions with those guys.

    Manton: And yeah, good updates to Xcode. No problems. But yeah, totally ignored anything with code generation as far as I can tell.

    Daniel: Yeah. And real quickly, I’ll just say, I also said something about this. I said, oh, well, you know, sounds like they didn’t say anything about AI.

    Daniel: And a few people chimed in, of course, with: well, of course they did. Like you said, the whatever mechanism transformer, blah, blah, blah.

    Daniel: That’s not what we mean right now when we say AI. Right now, when anybody in this world we live in, who is not like an AI researcher says AI, they mean large language models and the way that they provide chat or completion based services to users.

    Manton: Yeah. Assistant-like stuff and the way Apple is using, I think the transformer language model they were talking about, of course that is part of like chat GPT and whatnot. But I think they were talking about better auto-complete and behind the scenes, seamless stuff. Great. Fantastic.

    Manton: But that is just not the same thing as assistants that help you and that feel intelligent, even if they aren’t.

    Daniel: Right.

    Manton: All right, headset. Vision Pro.

    Manton: I was simultaneously blown away by this, and also I still really question the strategy. But from everything I can tell — I was listening to Dithering and Ben and John were just like, this exceeds our expectations of how technically good it is.

    Manton: And so I’m trying to take a step back and kind of separate, did they nail this versus should they have bothered to build this?

    Daniel: I’ve been joking for a while now, and I mean, the joke has totally come to a head with this idea that the reality distortion field is actually the product now.

    Daniel: It’s hard to know, and I agreed with you, I was watching the keynote and everything is so stylized because obviously they can’t show you, or maybe they could, they definitely could show you what it looked like just from the point of view of the wearer. I’m sure there’s like a screen capture recording released in their debug builds right now.

    Daniel: But obviously the way they presented it was kind of like giving you an idea by showing you video of a person who’s allegedly looking at the stuff.

    Daniel: I actually had the thought during the keynote, wouldn’t it be cool if the way they made those videos was one person wearing a headset looking at a room that happened to have another person wearing the headset in it?

    Manton: Welcome to the future, everybody’s wearing headsets, even when they’re in the same room with real people.

    Daniel: Well, and that gets at the real… To be honest, when I say my version of wrapping my head around things has a lot to do with the social implications of what such a product means.

    Daniel: I had the same gut reaction with Google Glass 10 years ago. You know, products that draw the user inward do not make me feel comfortable about them as social interactive products. You know what I mean?

    Manton: Mmm hmm.

    Daniel: And I mean, I guess to be honest, I wear my AirPods all the time everywhere. And I know there are people out there still who believe it is rude to be in public and to have headphones on, which I don’t agree with that assessment.

    Daniel: And so that might be like kind of an instructive example of people have different takes. But my take is if you’re going to block your eyes out to the world, do not talk to me. Do not talk to me.

    Manton: Haha. Yeah.

    Daniel: And even with my AirPods, I have social standards, I take my AirPods out when I’m going to like a retail counter.

    Manton: For sure. Yeah.

    Daniel: And I find it very rude if people don’t do that. I forgive it if they at least turn off the volume and the person doesn’t know that, you know, I kind of forgive it. But the case of something literally covering your face, I don’t want anybody interacting with me like that.

    Manton: No…

    Daniel: And I definitely do not want anybody interacting with their kids during like pivotal moments of their childhood.

    Manton: Yeah. And people will push back maybe on that and say like, well, at the birthday party, we all have our phones out anyway, taking video. It’s just not the same thing. It is not the same thing.

    Manton: And I have missed things because I’m looking at my stupid screen. It’s just not the same as covering your face.

    Daniel: I think it’s so instinctive to us that covering your face is a way of, I mean, like we do it as humans. We do it when we’re scared of something or we don’t want to engage with something.

    Daniel: And I think there’s a lot of instinctive, like animal response to this idea that somebody’s face is covered. And the eyes showing through is very impressive, but very not enough for me.

    Manton: Yeah. It was really impressive in the keynote, but I do wonder in real life if it’s just creepy.

    Manton: And I wanna acknowledge that we’re judging a product that we haven’t seen. Like some people have actually tried it, friends of ours have tried it, and their opinion goes a long way. We’re just reacting based on what we’ve seen in the keynote and heard people talk and blog about.

    Manton: But even putting that aside, I kind of went into the keynote thinking, if it’s technically amazing, would I still want to use it? And would I still develop for it?

    Manton: And it turns out like it’s even better than I thought. Even my wildest like expectations of Apple just completely getting this right for what they’re trying to do, it’s even better than that. Super impressive.

    Manton: And yet there’s something wrong about this product. And we’ve talked about like AR versus VR, mixed reality, etc. And I even more strongly feel now that I’ve seen it that the number of years I have in my head is 20 years.

    Manton: It’s gonna take 20 years for this product to be more seamless with actual glasses-looking things that you can actually see through and people can see your real eyes. More like real AR, at least 20 years, if it ever happens.

    Manton: And so it’s certainly a bet on the future, but it’s so far away. It’s so far from being a product that I could use in public, that I could use anywhere that there are people.

    Daniel: Yeah.

    Manton: I think that is an issue for me.

    Manton: I was thinking about this last night, so I’m in Colorado. I’m going to campsites with this van that I rented and just seeing nature and whatnot. And last night I made dinner. I sat at my like picnic table at my campsite and I pulled out my laptop, popped open a can of wine and I caught up on some work.

    Daniel: Haha.

    Manton: I’m chilling out, kind of in nature, but there are other people around. And I thought, I’m working on my Mac. Would I put a headset on here? No, I wouldn’t. It would be so weird.

    Daniel: Even in a fairly remote like.

    Manton: Yeah. Because people might walk by, drive by, it would just be weird.

    Daniel: You would be so self-conscious.

    Manton: Yeah, I wouldn’t be able to concentrate. It wouldn’t matter how much better the virtual screen is than my 16-inch, old-fashioned screen.

    Manton: And I think there are areas like that. I work in coffee shops every day. I was working at a coffee shop here outside Denver this morning for a couple hours. I just can’t imagine this product ever being used in that environment.

    Manton: Like I said, in 20 years, maybe it’s more like AR and more like real glasses, but this device is so powerful. It’s more powerful than my own Mac. To get it to the point that it can be seamless around people, man, it just seems so far off. I don’t know.

    Daniel: I think I’m not quite as pessimistic about the timeline that this technology stands a chance.

    Daniel: It’s funny because if you compare it to some Apple products, the chances are not great that it advances very quickly. For instance, the iPad was released 13 years ago. So if it advances at the rate, and just to be clear, the iPad today is an amazing device, so much more advanced than the original, but on the whole, it hasn’t changed that much.

    Daniel: The usage is the same, basically. And I think a lot of people expected a lot more development on that front.

    Daniel: On the other hand, something like the iPod is about 20 years old and is now essentially something that you wear in your ear. And that is the kind, so just to say, yeah, 20 years, maybe you’re right in that respect. But I could see it happening a lot faster because the unique technological angle here is going to fund this thing.

    Daniel: I think even with a very modest amount of buyers, even as a loss product, even if Apple were losing money, I think staying ahead of the pack on this type of product is going to pay dividends for Apple. And the saving grace for me is that I do think there is a point at which it’s kind of the difference between the AirPods being a helmet you wear over your head or the little device you pop in your ear.

    Daniel: I still feel a little self-conscious sometimes about my AirPods, but if it were a helmet that I had to wear, you know? Like, this is the helmet, and we’re going to see, you know, assuming everything goes the way I expect, we’re going to see, I think, a lot sooner than 20 years, the equivalent of the AirPod of this.

    Manton: Yeah, I dunno. Okay, but let’s think about that, though. The technological leap from wired headphones to the AirPods, I mean, it is significant. There’s, you know, cool chips in the AirPods that allow it to do what it does, but it’s not the kind of leap we’re talking about from a $3500 headset that has 12 cameras to something that looks more like glasses that you could wear in public, even if they’re awkward, bulky glasses.

    Manton: I’m really having trouble with that leap. I don’t see how you can do it. And think about how far the iPhone has come from gen one, couldn’t take video, to today. The iPhone’s great, it has come a long way, but even that is not the leap that we’re talking about, or I’m talking about.

    Manton: That is a very natural year-by-year evolution, and that’s their core product, where they sell hundreds of millions of these things. They’re not going to sell very many of these. They’re not going to update it every year. I’d be surprised if they update the headset more often than every three years. So we’re talking five or six generations. I don’t know.

    Manton: Man, I hate to be so negative. I just don’t think this is a product that Apple needed to build, or that I personally am gonna be into. I don’t know. It’s cool, though. It’s really cool.

    Daniel: It is, well, and like you, me hearing the dithering, John Gruber and Ben Thompson describing it as not being able to distinguish the edge of the quote-unquote screen. That kind of was a wake-up call for me.

    Daniel: There is no product, I think, I think even all the other VR headsets, I don’t think there is a product in the world. The combination of like, whatever Apple did to make the thing sit so close to your eye, and the fact that it’s such a high resolution, there’s no product in the world where you can kind of fool your mind into thinking you’re looking at reality.

    Daniel: And that’s why the whole reality kit, whatever, sort of all makes sense now.

    Manton: And really, every time they say, “only Apple could combine blah, blah, blah”, I always kind of roll my eyes a little bit. It’s totally accurate in this case. There is no company that could have built this.

    Daniel: Nobody else has the money to do the research and the social influence to make it, you know, even vaguely. Like some random little like company headed by a billionaire could have come up with something like this, and it wouldn’t have gotten traction because it wouldn’t have captured the public’s imagination.

    Manton: Right, and only Apple, they have all the right skills to pull this off. And they don’t have a great track record of hobby products improving. I would say none of their hobby products have really improved very quickly, which makes sense. Their business is built around the iPhone and the Mac. Apple TV is more or less the same as it was, you know, a dozen years ago.

    Manton: And that’s fine, but they obviously care a lot about this product and they could change that reputation. They could have this fringe side product that they pump millions and millions, billions of dollars into, and they’re patient.

    Manton: Again, they wait 20 years and then you have the Vision Pro and you also have the Vision Air or the Vision Lite or whatever you want to call it, a more mainstream device that is not as bulky, not as powerful, something that more like normal people maybe would be into. That could totally work. It feels really far off. I’m gonna be old or dead by the time this happens.

    Daniel: I think there’s two things that are remarkable about this product. One that struck me right away was how much of an amalgamation of other Apple technologies it is in one unit. And you think about all the things they’ve been working on.

    Daniel: The little digital crowd is just like the Apple Watch. The cameras actually remind me of whatever research they may have been doing into self-driving cars. You have super high resolution screens. The audio playback in the device is also using that kind of room reading, LIDAR type stuff.

    Daniel: That’s one aspect of how “only Apple” could do this because they had all these other products in the works that sort of contributed to this device. But I also think this device is, regardless of its commercial success, it is a technological test bed that can easily spin off other devices that benefit from the advances they make with this.

    Manton: Yep. And I can definitely buy the argument that someone had to build this. That’s not true, but there’s value in building this purely for the research and technology and just to see what’s possible. Because if no company pushes the edge and tries new things, we don’t get anywhere. There is a role for pushing, just seeing what we can do, whether it’s a compelling product or not.

    Manton: So yeah, I totally respect that. And I think the people that worked on this should be very proud of what they’ve built. It really does look incredible. It’s just a question of, strategically, is this something that Apple should do? Is it good for humanity? And then secondarily, for developers, is this something that should be part of our strategy or not?

    Daniel: Right.

    Manton: If this is the — again, 20 years in the future — this is the replacement for the Mac, is this something that we should be fiddling with and paying attention to and watching all the WWDC sessions and playing with the simulator. I’m curious what you feel as a mostly Mac developer, how much are you going to pay attention to this?

    Daniel: Yeah. I mean, it’s so much to take in because as you saw during the keynote, there’s like elements of how it, how it stands to just like integrate with your Mac and show you your Mac desktop apps on like virtual monitors within the device.

    Daniel: And then there’s the other angle of native apps for this thing, maybe subsuming the uses of traditional Mac software, iOS software, all the above. So I’m going to keep an eye on it. I do, my instinct is to mostly ignore everything, especially since there’s no SDK yet. And I find it kind of awkward to even consider jumping into WWDC sessions about this when there’s no hands-on opportunity to do anything, even with a simulator. So I’m going to keep a cautious eye on it.

    Daniel: I think there is something to what, I mean, you know me, like both of us have been pretty dismissive of AR VR the past several years. And I think the way we’re talking about this still reflects our skepticism about it, but it brings it into the real world in a way that I don’t think is going away.

    Daniel: Having a point your iPhone at a table AR experience never felt compelling beyond the slightest novelty. But let’s just assume that this thing didn’t cost $3,500 and it costs $350, which might be the case in five years, 10 years.

    Manton: 20. I’m sticking to 20 years.

    Daniel: Haha. Well, the situation will be different when there’s a massive market of people who have these things.

    Daniel: And I just want to say one more like slightly optimistic thing. I think my reticence and spooked-outness about this device is somewhat to blame on Apple’s presentation of it during the keynote. The way that they tried to frame it as something in particular that you would wear, like I mentioned, like when you’re engaged with your kids during special moments, that’s the opposite of what I think this type of device should be used for.

    Daniel: And if I focus on the things that it’s good at… I won’t fault any parent for wearing a welding helmet while they’re building, you know, a rocking horse for their kid. That’s a special moment in a family’s history as well.

    Daniel: But let’s not try to conflate the wearing the welding helmet with spending special time with your kids. Likewise, I think there are a lot of parallels actually with this to a device like a television, because televisions are famously attention absorbing. You can’t really talk to somebody who’s watching a television. And frankly, in my opinion, having a television on in any kind of social environment, unless you’re like watching a sports game with somebody is not conducive to sports or to social interactions.

    Daniel: But I think if I try to look at this more generously, as like, it’s another kind of TV, or it’s another kind of welding helmet. You use it on your own time for your own purposes to either entertain yourself or achieve specific productivity goals. Then a lot of my negativity about it goes away. It’s just the idea of it being integrated as an everyday piece.

    Daniel: It’s the same idea as if I envisioned everybody walking around with a television hanging off their forehead. No, we, nobody really wants to live in that world. So don’t highlight that as the use case for this device.

    Manton: Yeah, I agree with you. I think that was a bit of a miss. Overall, I think the keynote will go down as the best in years, but I feel like some of those parts of the, you know, advertisement video that we watched, basically, we’ll remember them a little bit like the watch Apple Watch introduction. There was a lot of sort of gimmicky features. You know, texting your friend in class or sending pictures.

    Daniel: Sending your heartbeat to somebody.

    Manton: Exactly. I think some of these things will go down in that light. I love the Apple watch. I’m wearing mine now, I wear it all the time. But I think they were a little bit confused about their message when they introduced the watch and they sorted it out.

    Manton: And I think this is going to be the same way. I think we’ll look back on some of those and be like, no, we’re not putting this headset on for a birthday party. That’s not going to be right.

    Manton: And really, that’s why I kind of think about just strategically. I mean, we already have this problem that we didn’t have 15 years ago, which is that the family is sitting around TV, watching a movie or something and all the kids are on TikTok. This is already an issue.

    Daniel: Yeah.

    Manton: I’m not interested in it getting worse by actually having people close their whole face off from the world. So I think there are really legitimate uses for this, but in my life, they are very small.

    Daniel: Yeah.

    Manton: I like to work out of the house, I like to go places. I mean, I’m kind of an introvert. I don’t meet a lot of people and hang out with people in person. But when I do, when I’m around other people, it feels good to be around humans. We need that.

    Manton: And just strategically, I guess I’m really pessimistic about it. I think maybe it’s okay for us to have those two thoughts in our head of just technically and pushing the envelope, this is a huge breakthrough. On the other hand, as a human and as a developer, I’m fine. Zig when they zag. If everybody else is obsessed with this and WWDC over the summer, I’m fine focusing on other things and just continuing working on Micro.blog and working on these other aspects of social networking.

    Daniel: Yeah, I agree. In a sense, I am kind of relieved that I don’t feel like I’m going to have my work cut out for me in terms of new Apple technology over the summer. None of this is like, oh, drop everything.

    Daniel: The one product that I think might have a little bit of a compelling case is like Black Ink with the idea of maybe like solving a puzzle on a virtual sheet or something. Even that’s a little bit of a stretch, but you know, maybe against a flat desk or something. But I don’t know. Obviously I have other Black Inks to ship before I get to that.

    Manton: Yeah. I want to ask you, because I don’t know if you saw Apple News+, I think has some crossword thing.

    Daniel: Oh yeah. Yeah, I did see that.

    Manton: We’ll probably have a whole show next week about everything non-headset related. There’s really a lot at WWDC and we’ll have to tackle that next week. Because the headset is just the obsession right now.

    Daniel: Yep, agreed. All right. Well, I hope you continue to have a great trip. I’m glad that this time around we’re recording, you’re in your van again. If folks remember the last time Manton recorded in his van, it was like a rainstorm.

    Manton: Yeah. And unfortunately… So I bought a new mic. That’s a whole ‘nother story about going to three different Best Buys, trying to find something that I could take with me. I did a test recording and there’s a lot of background noise still. So apologies, everybody. Hopefully it turns out okay.

    Manton: But thanks everyone for listening this week. Enjoy WWDC, and we’ll see you in the metaverse. I dunno.

    Daniel: See you in my glasses.

    → 8:32 AM, Jun 11
  • I don’t think Apple needs to obsessively avoid saying AI. “Machine learning”, “transformer-based language model”… It just sounds awkward, but I get why they do it.

    → 10:41 AM, Jun 5
  • I’m always fascinated when our new AI bot overlords get really basic facts wrong. They know the answer, but can’t put it together until you press them on it. Here’s a quick transcript from ChatGPT that I ran into today.

    → 7:58 AM, Apr 26
  • If it was starting to feel like Humane’s device would be overhyped or vaporware, I’m ready to put aside those concerns. It looks like they are onto something fascinating, both the projection and the AI language translation.

    → 1:03 PM, Apr 21
    Also on Bluesky
  • There’s a lot of talk about how AI can get facts wrong. That’s fair, but in my experience it’s correct most of the time. Even when it’s slightly off, there’s usually some useful truth in the answer. Much more frustrating is voice assistants who can’t even begin to give an answer.

    → 10:00 AM, Apr 13
  • Fascinating discussion on the latest The Talk Show with @gruber@mastodon.social and @danielpunkass about AI. What does it mean for humanity when more tasks can be automated and fewer people need traditional jobs? As a developer, a lot of my time isn’t coding but thinking how something should work.

    → 12:30 PM, Apr 6
  • Posted Core Intuition 551! This episode covers the launch of Micro.blog’s new podcast transcripts feature, and more generally what AI is good for. A short show, just 22 minutes this week.

    → 12:11 PM, Mar 31
  • Introducing Micro.blog podcast transcripts

    We’ve launched a new feature for Micro.blog Premium customers: automatic podcast episode transcripts, powered by OpenAI’s Whisper model. I’m excited about this because it’s one of the more practical, time-saving solutions coming out of the rise of AI. The automatic transcripts are so accurate they can be used as-is, or edited by hand as you have time.

    I thought it would be clever to ask ChatGPT to write a blog post announcing this feature. You can see the result here. But I threw it out because I like writing my own blog posts!

    At Micro.blog, we don’t usually reach for automation first. We curate the Discover section by hand, looking for posts that will provide a snapshot of activity on Micro.blog to help you find new people to follow. We don’t have trends and don’t have public likes or retweets. AI is going to reshuffle many tech products, but we’re never going to have an AI-driven algorithmic timeline. AI is a tool that we’ll only use when there is real benefit that aligns with our principles.

    The new transcripts feature is available to anyone hosting their podcast on Micro.blog. When you upload a new MP3, Micro.blog will process it to generate a transcript. You can then edit the transcript or link it from your podcast page.

    There is also a new Micro.blog plug-in to add a list of transcripts to your blog, as well as control options to disable transcription and link to the transcript from blog post. Search in the plug-in directory for “Transcripts” to install it.

    There’s no extra charge for any of this. Micro.blog Premium has always been $10/month and it will continue to be priced that way. It includes podcast hosting and also email newsletters, bookmark archiving, web page highlights, and much more. We think it’s a great value.

    → 6:02 AM, Mar 30
  • Watching some of the recent Microsoft Copilot demos and playing with Bing, I wonder what AI products they will roll out before everyone else can even attempt to catch up. Maybe a Microsoft voice assistant box? It would leapfrog Alexa and Siri.

    → 2:16 PM, Mar 20
  • Loish on AI art

    I haven’t been able to pin down exactly what I don’t like about AI-generated text and art, but I always got the sense that something wasn’t right about it. I like this post from Lois van Baarle, writing about artists protesting AI:

    Many have compared image generators to human artists seeking out inspiration. Those two are not the same. My art is literally being fed into these generators through the datasets, and spat back out of a program that has no inherent sense of what is respectful to artists. As long as my art is literally integrated into the system used to create the images, it is commercial use of my art without my consent.

    She’s an incredible artist with a very distinct style. If it was copied by a robot, you’d know. Go check out her Tumblr blog or books.

    → 1:28 PM, Dec 15
  • This week’s Core Intuition is up with discussion of the MarsEdit 5 release and ChatGPT. We go long on this one, about an hour, possibly because I can’t stop rambling incoherently about AI.

    → 1:58 PM, Dec 9
  • AI is getting too good.

    → 1:06 PM, Dec 2
  • Core Intuition 248

    This week on Core Intuition, Daniel and I talked about recent Apple news:

    Daniel and Manton react to the European Union’s €13B retroactive tax demand to Apple, talk about the impact of tax laws on indies and small companies, and weigh in on Apple’s purported AI and machine learning triumphs. Finally they catch up on their ambitions to be more productive as the busy summer transitions to fall.

    I wondered whether Apple is so obsessed with privacy that they are blinded to what is possible with more computation and extensibility in the cloud. I judge their efforts not only by the remarkable work the Siri team has done, and by what Google and Amazon are building, but also by Apple’s own gold standard: the Knowledge Navigator video from 1987. That vision is too ambitious for one company to develop all the pieces for. We eventually need a more open Siri platform to get us there.

    → 9:57 AM, Sep 1
  • RSS
  • JSON Feed
  • Surprise me!
  • Tweets