pull down to refresh

There's a good argument to be made that AGI doesn't exist. Even these current state of the art models are little more than very accurate autocorrect.
I'd be very interested in hearing such an argument as to why AGI does not exist, even 100 years into the future. I'm not being facetious, I have no strong opinion on it.
Indeed, in my experience, it is powerful, but when asked to work on very specific tasks, but I wonder if there is a stronger argument than just my anecdotical evidence or people's intuition. For instance, I never would have thought that the early crappy chat bots from 10 years ago would become this powerful this fast, but here we are.
reply
100 sats \ 1 reply \ @Zk2u 11 Jun
Of course, me neither! However, I think we'll need something different to build AGI other than transformers, which means we're back to the beginning before the AI craze.
It's worth saying degenerative models (lol) are still black boxes mostly to the researchers who build and train them. They're still not quite sure how they make the connections they do. It certainly is powerful and I also use them very often now, however LLMs are very grounded in reality rather than science fiction.
I don't know quite what I'm saying now 😂
reply
However, I think we'll need something different to build AGI other than transformers, which means we're back to the beginning before the AI craze.
As it is all quite speculative indeed, this is good enough for me. I did not believe in AGI coming any time soon before the existence of these transformers, so assuming those won't be able to do the trick (and from my understanding superficially studying them, they are indeed not able to do anything truly creative), I can rest assured it's not going to happen any time soon.
reply