I think there’s a happy medium between doing nothing and burning cash faster than the US military like OpenAI. If I had to pick one company who walking that line the best, it’s Google.

You can wait until improving hardware eventually solves the local LLM problem but imho that’s too passive.

What if someone cracks the problem of splitting LLM inference effectively between local device and server? Think about it. ChatGPT can do calculus. Is that useful to most people? No. Can you currently effectively modularize an LLM and load knowledge on demand? No.

I’m fairly bearish on the use cases for current AI. The biggest is actually just firing people and suppressing labor costs.

But a personas assistant, at least in theory, is something people want, even if it’s just to effectively obey voice commands. If Apple loses to Google here it’s going to be bad for Apple. I think they have to do more than they’re seemingly doing.