@futurebird I agree that current LLMs are not conscious. But nor are they simple Markov chain text generators - are you familiar with Anthropic's work on transformer circuits? Plus, "current approaches" includes hybrid systems like AlphaGeometry which combine neural networks and symbolic theorem provers. Like I said, I don't think we'll get to AGI simply by iterating on what we have now. But I didn't think we'd see an AI get an IMO gold medal this soon either.
P
pozorvlak@mathstodon.xyz
@pozorvlak@mathstodon.xyz
A forum for discussing and organizing recreational softball and baseball games and leagues in the greater Halifax area.
Posts
-
ugh I remember this mf from 90s usenet, he would pontificate endlessly but never seemed to actually work on anything -
ugh I remember this mf from 90s usenet, he would pontificate endlessly but never seemed to actually work on anything@glitzersachen "current approaches won't scale to ASI" seems plausible (though not so plausible I want to bet the farm on it), but you totally lost me at "...and then there will be a fifty-year AI winter". I give it five years max after the current AI bubble bursts before the next one starts inflating.