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.

Train bridge at Falling Waters, West Virginia. On the other side of the river, the Confederates were stranded for days after Gettysburg.

So with threads.net switching to .com, I was a little worried about how this might impact the fediverse support. Thankfully looks like they’re sticking with .net for handles.

Roanoke River overlook, along the Blue Ridge Parkway.

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.

Had the Switch 2 pre-order in my cart but kept changing my mind, gonna skip it until there’s a game I really want. Hopefully won’t be too hard to get later. 🕹️

Nashville Public Library.

I really like this post from Ashley Willis, about doubting yourself and the fear of what loud people on the internet might think of your writing:

I don’t know exactly when that changed. There wasn’t one big moment, just a slow fade. Something dimmed. I started second-guessing myself more.

Micro.blog’s backend is running so much more smoothly now after I addressed some memory issues yesterday. I often assume I know where problems are, and it takes actually digging in to discover I was wrong, the problem is fixable in a different way than I expected.

Nailed it. I’m too amused by wall outlets and light switches gone wrong, it’s surprisingly common. This one in the hotel stairway.

I’ll be speaking at EFF-Austin next month! An updated talk about blogs, social networks, the fediverse, and where I think the open web is going.

In the Bluesky announcement about checkmark verification is the tidbit that 270k accounts have linked their username and domain name. 270k custom domains! That is very cool and a great sign for the open web.

Micro.blog went off the rails today, for some reason the day I’m out of town an old memory leak decided to blow up into a much worse problem. I think I’ve got it under control now.

Speaking of Objective-C, when I see example code that looks like this:

NSMutableArray<NSString *>* paths

I simplify it to:

NSMutableArray* paths

I don’t think declaring types everywhere improved the language at all. Let a dynamic language be dynamic. 🤪

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.

There’s no place I appreciate Apple Pay more than at the gas station air pump, instead of using quarters. 🛞

Yair Rosenberg writing at The Atlantic about the arson at the Pennsylvania governor’s residence. The suspect was struggling with his mental health and influenced by attempts to vilify Josh Shapiro over the war in Gaza:

But those struggling with internal demons don’t originate our external ones; they reflect them. In their confusion and pain, such individuals latch on to those already targeted by the broader culture and its preexisting pathologies, showing us not who they are, but who we are.

New laptop sticker, from Vintage bookstore and wine bar. Booksellers not algorithms. 🍷

Thinking about AGI. The big step that is missing is personal AI being able to learn when it answers a question. So if I use deep research and my AI goes off and spends 10 minutes researching an answer, all of that should be fed back into the model for later.

Gus Mueller:

To me, Objective-C has always felt expressive and capable, doubly so when I first started using it.

After Swift became popular, I felt kind of guilty still using Objective-C so heavily, but I’m over it. Micro.blog for Mac is all Objective-C. I did two new releases this week.

Finished upgrading a server. Some things should be faster! (And some won’t be.) I’ll continue to look for places to optimize.

Coffee yesterday at Lazarus. ☕️

Now that I’m using Hetzner in the EU, I’m having difficulty understanding Linode pricing. For example, dedicated 16 CPUs on both hosts:

Linode: $288
Hetzner: $110

This is a massive difference. Is Linode that much better? I feel like a fool for paying this.

Micro.blog for Mac continues to improve. Just released version 3.5. A new feature I like in this release: paste a photo from the clipboard directly into the Uploads section to upload it.

Also added a Preview button. I use this with command-shift-P but I’m sure some people didn’t know it was there.