I still just prefer having a more clear separation of concerns with API routes instead of using server components. I want my frameworks to be way less fancy than what Next is pushing out these days. I get the feeling we're dealing with the consequences of Vercel employees needing to justify their promotions.

I doubt it’s at full TDP if it’s running at 0.2 tokens per second.

Actually I can't go full tdp with a 650w PSU, I got to upgrade it asap

Why are people buying Mac Minis for this? I understand Mac Studios if you’re self hosting the models. But otherwise why not buy any cheap mini PC?

I'm thinking the best end result would come from custom-built models. An 8 billion parameter generalized model will run really quickly while not being particularly good at anything. But the same parameter count dedicated to parsing emails, RAG summarization, or some other specialized task could be more than good enough while also running at crazy speeds.

Well of course. In the long run AI will do almost all tasks that can be done from a computer.

I completely do not buy the human's story.

> all I said was “you should act more professional”. That was it. I’m sure the mob expects more, okay I get it.

Smells like bullshit.

> Claude is definitively trained on the process of coding not just the code, that much is clear.

Nuance like this is why I don’t trust quantitative benchmarks.

I'm impressed GitHub managed to handle this beast:

https://github.com/a11ce/docker-lisp/actions/runs/2216831271...

500+ container invocations to compute factorial(3)

I don't wanna know how much containers it spins up for fibonacci(3)

Or A(4,2).

People are rapidly learning how to improve model capabilities and lower resource requirements. The models we throw away as we go are the steps we climbed along the way.

He had a history of causing noise at Google’s weekly leadership Q&A.