Vent: A developer at a meetup last month said AI progress is basically done

I was at a small tech meetup in Portland, and this guy from a local startup was going on about how large language models have hit a wall. He said, 'We're just making them bigger for tiny gains, and the real innovation stopped last year.' He was so sure of himself, but then he admitted his own team's model still hallucinates basic facts about 15% of the time. How can you call it done when the core problems aren't even solved? It felt like he was just trying to sound smart while ignoring all the new research papers coming out weekly. Has anyone else run into people who are weirdly eager to declare the whole field finished?

3 comments

3 Comments

xenarobinson22d ago

Wait, his OWN model still gets basic facts wrong 15% of the time and he says progress is DONE? That's like saying you're done building a car when the wheels fall off every few miles. New stuff is coming out all the time, like models that can actually reason through math problems step by step now. How can you look at that and say we've hit a wall? It just sounds like he gave up.

ryan71922d ago

But is getting facts wrong 15% of the time actually a big deal for most things people use it for? It's not like we're perfect either. The new reasoning stuff is cool but it's still just a different kind of pattern matching, not real understanding. Maybe he just means the easy gains from making models bigger are over, not that all progress is done.

alices1622d ago

Forget the 15% thing, that's not the real issue. The problem is calling it "done" when it still messes up basic stuff. Feels like giving up right when things get hard.