2026-07-28 01:50 UTC
@lzg@mastodon.social For me, honestly, no. They work on the same principle that allows us to readily see faces in pieces of toast, but it's still just toast. Right and wrong are the same for them. The fundamental issue is the lack of world model, hence why they can't play chess: after a short while, they start making illegal or nonsense moves. They can repeat the rules of chess but can't apply them to the board because to them there is no board. Or... world. We've other far better repositories of human culture.
Replies (2)
-
@lzg@mastodon.social 2026-07-28 01:53
@conniptions@mastodon.social yes of course we have better things. but LLMs exist.
-
@Rakun@mastodon.online 2026-07-28 07:34
@conniptions@mastodon.social @lzg@mastodon.social yes, and to be precise, modern chatbots massage the text quite a lot so that recognized patterns are sent to specialized tools. That's why they suddenly become able to do basic math or count the Rs in strawberry... sometimes As far as I know, the LLMs at the core of these chatbots remain precisely fancy autocompletes.