Hacker News Reader

Comment by encoderer | Hacker News Reader

LLMs are amazing but my gut says they’re going to plateau. No superintelligence.

I suspect this is a contrarian view but I’m not sure?

I am fairly confident LLMs are plateauing or will do so soon. I am also fairly confident that there are other kinds of models, either already invented or yet to be conceived, that will go further than LLMs.

I don't think this is controversial at all.

My impression is that this is the most common view. I also think it's the view that's most likely to be correct, though, so I may have a bit of selection bias here.

It is certainly contrary to the view held by some subsets of people, though.

i don't find it to be that common. even the attitude of "just wait for gpt5" seems prevalent. why is it to obvious there's a new amazing version around the corner, which replicates the jump between gpt3 and 4?

I find it to be the most common view of people who aren't that involved, monetarily or emotionally, with the AI industry and especially among people who aren't that involved with the software industry in general.

Here on HN, the people who think that this stuff is the start of some sort of whole new world are overrepresented, as we would expect. HN readers tend to be a fairly specialized subset of even the software industry.

> why is it to obvious there's a new amazing version around the corner

I'm not sure what your question is here. Are you asking why it's obvious, or why it's not obvious? My bias is that I don't see much reason to think that LLMs are on the edge of some kind of massive leap into AGI, let alone a "superintelligence".

> Are you asking why it's obvious, or why it's not obvious?

you're right it wasn't clear, that was rhetorical not aimed at you

I think agents (not the gimmicky ones) will do a lot of interesting stuff in the future, particularly where research is concerned.

At the moment, the big limitation is context, memory and compute.

But I do think there’s a lot of temptation in having a black box AI that bangs its head against a wall until it finally brute forces a solution.

Well, it's a very tentative, but moderately sized step on the loooong ladder towards ASI. The thing is, there is an unknown point on the ladder where the end goal could start taking the steps itself.

Does ASI mean artificial super intelligence or something else?

Yep. ANI is what we have now (narrow), AGI is what is next (general, humanlike) and there are theories that when we have that, ASI (super) is just around the corner.

At that point we will either blip out of existence or transcend to a higher plane of existence and become a possible Fermi paradox explanation.

You might want to define your acronyms until we have AI to help us understand.

Amazing for fiction, sure.

Reply on news.ycombinator.com