What is smarter? If smart is doing infinite calculations at lightning speed then AI passed that decades ago. In many areas, AI is smarter than any individual so saying that AI will be smarter next year means nothing. Let's define the mark that defines smart first.
Almost all of the top models are smarter than 99% of the people already I believe. They speak a lot of languages, and they are pretty proficient in almost all areas. Yes, people can be smarter than the AI in the areas they specialize in, but generally speaking llm’s know all subjects and almost everything else. They are only going to get better while I don’t think we humans will get that much better, maybe in one or two areas but not everything all at once.
If we’re talking about knowing existing facts about the world, absolutely.
They’re also better than most of us at creating art (rhymes, poetry, digital paintings, etc) that falls into existing clearly defined patterns.
But when it comes to deep creative thinking, inferring new knowledge and very long chains of logical reasoning, I’m not convinced they’re actually anywhere near human intelligence yet.
I think we’re missing a key algorithm for true general intelligence.
I suspect this algorithm will involve a feedback mechanism that will be incredibly hard to tune and get exactly right. We already see that the relatively simple LLMs we have today can go “crazy” fairly easily if the parameters aren’t tuned quite right. This will get exponentially harder if the AI is undergoing constant self-modification.
It’s possible that “AI psychologist” will become a common profession in the future, and I suspect it won’t be significantly easier to make progress in that field than it has been for humans.
But when it comes to deep creative thinking, inferring new knowledge and very long chains of logical reasoning
I haven't found humans to be all that good at it, either. We do it sometimes, under limited circumstances. A few people are creative geniuses.
But most people just don't. An average chain of logical reasoning is perhaps two steps deep. Mozart is Mozart and Taylor Swift is Taylor Swift but I can't reliably play a scale on a guitar, and most people can't either.
Humans are still a lot better at doing ordinary human things: walking, figuring out how to do things by watching, guessing what's on your mind by looking. But the really "deep" stuff is actually kinda rare. Maybe we're still better at it, but I'm not sure we're all that good.
Agree, and this is what Sam Altman says on this topic:
"If superintelligence can't discover novel physics, I don't think it's a superintelligence. And teaching it to clone the behavior of humans and human text - I don't think that's going to get there."
> Almost all of the top models are smarter than 99% of the people
Not true. Most humans can learn driving in under a week. How many AI models today can learn driving in under a week, just like a human? None. Until it can, AIs are not smarter than humans.
They can’t understand the world like we do, but on the knowledge scale they know more and they can reason pretty well based on the knowledge. Its not AGI that can do anything generally, but being “smart” means IQ tests and SAT tests and stuff, they can do better than most humans.