Siri AI context
Instead of thinking about on-device models and cloud models, a better framing for Apple vs. everyone else is on-device context and cloud context. Apple is focused on apps, privacy, and security, at the expense of speed and universal access anywhere.
There can’t really be a web version of Siri AI. There can’t be a fully-featured HomePod version either, except in limited areas where Apple could sync its own data. Siri will continue to be fragmented across platforms.
I personally want an assistant that knows a lot. Yesterday, I asked ChatGPT whether 7-Eleven carried a certain medicine. In its answer it included a map because it knew I was traveling, and it also included a minor health tip based on something I had told it last week about an x-ray. Some people will find this creepy and invasive. They’re not wrong. But I think it should be my choice whether I let my assistant know these things.
Apple’s strategy will resonate with a lot of people. Perhaps it is the best approach when you have billions of users. It doesn’t really excite me, though, because I can see what is possible when you instead put the context and memory in the cloud, where an ambient assistant and its helper agents can get to work for you.
I also might be wrong about the limitations. For example, Siri AI could manage memory in the same way that OpenClaw and Codex do, with on-device data that then syncs between platforms. In that world, memory could feel universal even if it is never stored in the cloud.
Apple should continue to do what they are doing. OpenAI and Google should continue to do what they are doing. Every approach is a trade-off. Apple reasonably trades ubiquity for privacy. Each company following its own path is the best way to give users choice and discover what works in practice.