I personally don’t think LLMs are anywhere near ready for this type of rollout, so it will be a while before we see results. The cost of something like GPT 3.5 deployed at scale is tremendous. The censoring of responses will make it an obstreperous assistant, unless this has gotten considerably better in the last few months. And are we all just going to accept hallucinated results over something that is either vetted or clearly misunderstood?
I don’t even know how far away we are from sufficiently powerful TPUs that fit a mobile case and power profile. But I do think this is likely to be the next big mobile advancement, however far off it may be.
Latest autocorrect is already based on LLMs. https://9to5mac.com/2023/11/14/ios-17-2-disable-inline-text-...
> just integrate the latest AI tech into them
The Ultra 2 watch has a GPU (neural engine) that can run on-device Siri. Works great, I've been able to run basic commands like setting timers and local shortcuts in areas with no cell reception.
They have had powerful devices with best-in-class integration for many years now, and Siri is still a frustrating experience and lags significantly behind competing voice assistants.
Competitors also have access to LLMs, and competing ecosystems are maturing (e.g. Samsung, Google). I wouldn’t be surprised if Apple makes waves with AI, but I also wouldn’t say they’re particularly more likely to do it than anybody else.
Now rather than deal with all this cruft, or deal with competing assistants (e.g. why the fuck does Google Assistant still exist if you have Bard???) Apple has the runway and the capital to make this type of LLM upgrade to Siri and the well-maintained ecosystem.
Because an integrated smart home assistant is a different problem domain than a general purpose chatbot.
ChatGPT was a revolutionary technology that released less than a year ago and surprised the world with its capabilities. It’s going to take some time to integrate that technology into an ever-present integrated assistant that leverages your personal knowledge graph to take actions on your behalf, but they’re working on it.
And Assistant imo has a huge leg up because of the investment in software integrations made in the pre-LLM world. Meanwhile when I ask Siri to show me photos of myself, it does an image search of the web and returns pictures of the word “myself”