Silicon Valley Daily

mid curve read of Apple LLM reasoning paper: haha, these guys can't even ship an LLM but they are criticizing everyone else 

S tier programmer who probably hates Apple more than anyone: yeah this is correct

Main takeaway from Apple's "The Illusion of Thinking" paper imho:

- The authors introduce a rigorous puzzle-based framework for quantifying reasoning ability.
- Beyond a critical complexity threshold, accuracy plummets to zero.

Read it yourself :)

https://t.co/AUtIGp7q6a https://t.co/gMG1mj4ftc

AI is in an awkward state of development. We have these tiny (4K) prompt embedding vectors capturing relations among abstract notions, connected to small (64K+) literal text or image contexts, and a static large (2T+) model. It’s enough for some magic but not true intelligence.

Tim Sweeney

mid curve read of Apple LLM reasoning paper: haha, these guys can't even ship an LLM but they are criticizing everyone else 

S tier programmer who probably hates Apple more than anyone: yeah this is correct

https://x.com/TimSweeneyEpic/status/1931802561826406653

Good retort on the paper if that's what you're into: https://www.seangoedecke.com/illusion-of-thinking/

IMO we don't have good goal posts on "true intelligence" and "reasoning", gotchas in either direction keep getting broken with another loop here, system prompt there, a fresh model, etc.

But as you said, none of this matters for Apple as a business, they'll be fine.