Posted by shioyama 11 hours ago
The big superhuman areas are speed, motivation (infinite), context size, and associative memory.
We are in the “singularity,” which will be a sigmoid, but we don’t know where it will reach its next level.
What does it not seem to have? True volition. Creativity too, maybe, depending on how you define that, which is a can of worms.
A calculator is superhuman. In speed of calculating numbers. Then again, Wikipedia lists calculators as AI, for just that reason.
I don't personally believe that intelligence is one dimensional, to the point of doubting that "general intelligence" exists or is even a valid concept. It's like saying "general locomotion." There is no such thing. There are wings for flying through air, fins and rotors for water, legs and wheels for land, rockets for space. A wheel doesn't work well in the air or space and a rocket is a poor locomotion method in the water.
This implies, BTW, that I don't think humans have general intelligence, because I doubt the existence of such a thing categorically. We get all up in our egos thinking we do because we feel so smart at all the things we're smart about, and of course we can't see the other things because we can't see them. Might not even be able to imagine them.
A chess algorithm is superhuman at playing chess. A human beats AI at writing prose or telling a joke. AI can beat a human at bug-hunting in a giant code base or constructing a large mathematical proof. Humans can beat an AI at dreaming up entirely new directions for mathematics. AI can beat humans at taking those new directions and fleshing them out and relating them to current art.
And so on.
We in fact tend to build machines to be hyper-good at things we are not good at, because that makes them useful to us. So we make AI that is better than us at fast thinking, huge context windows, huge scale associative memory, etc., just like we made classical algorithmic computers that were hyper-good at doing math.
There has been an open question in psychology for almost 100 years now: can you design a solvable abstract task for human performance that isn't correlated with this principal component? No one seems to be able to do it, and the AI models have only reinforced this. Making AI models smarter makes them better at almost EVERYTHING. The scaling properties of LLM models, and the scaling properties of neurons in the brain, ensure a general improvement of pattern matching that seems to be task invariant.
Personally, I am convinced by the speed hypothesis here. Since IQ in humans is very correlated with thinking speed, the ability to generate and evaluate more hypotheses per unit of time yields general improvement.
Also, this is an early stage technology that might be promising and it should probably be compared against x86, ARM, and RISC-V in their early stages. I feel like we are comparing a young up-start against established legends.
yeah this is why the "AI datacenter" debate frustrates me. People take it at face value that the enormous power demands of AI are immutable - they're thinking in a 20th century way where we're powering like, air conditioners or something that aren't changing. It's not at all a given that AI is going to demand power in a year or two the way it does now, these are problems that should be solved, not used as a reason to abandon the whole technology.
Is it that they don't want to overstate their claims to investors who have to decide whether to actually build these, in case the results they're predicting don't add up?
Either way, we need to be working on EDAI, so if they've got something, hopefully it's something.
We used to say ‘show me the code’, we need a bit more proof today. Or just wait it out to see if the authors are still pursuing the same idea 3 months later or have they already moved on.