Ian Bicking on Nostr: I've read a lot of discussion on #LLMs and whether they can reason. The more I think ...

I've read a lot of discussion on #LLMs and whether they can reason.

The more I think about it the more unreasonable the debate seems. Obviously LLMs can reason. Even simple instructions require reasoning... parsing instructions that use negatives, pronouns chained conditionals, this is all normal LLM capability. There's no pattern match against a training corpus that can accomplish that.

I think there's a kind of bias against normal human thought that dismisses this as "reasoning" ...