Of course they can't just talk to themselves and improve the knowledge of the world. They are just stochastics parrots that give you an average answer.
LLMs giving you something novel would be like if you let a model play chess against itself and become the best player in the world this way. Totally impossible.
So, the Jacobian Conjecture was done, via an counterexample. Can we now put the models to work on an even more difficult problem: Scrolling 100k of text, on a 128GB 24 core processor, smoothly in a browser?
Oh, but it will. Especially in math you depend on definitions and concepts defined earlier, and if you fall behind, perhaps in a class setting, it is hard to catch up, without having somebody who can explain it to you.
But, now, with world class knowledge possessing tutor, you can ask for an explanation of missed concepts, even embarrassingly stupid questions you would never ask a person.
Interestingly enough, ChatGPT started his answer like this only once:
"This is exactly the question I would ask next. My impression is:
Most standard invariants are...."
And this was a response to this prompt:
"Is there a chance of an indirect argument of X ~ A^3 coming from computing some invariant of X that forces it to be A^3? (I am not all that expert in algebraic geometry but I'm thinking like degree or Betti numbers or something.)"
Hey, Mr. Tao, where would you put the intelligence of the current models compared to top humans you surely mush have worked with? Perhaps given in the count of people you know that are higher than the newest LLMs? Also, how would you rate the speed of work compared to what your speed is, if this makes any sense?
Nice list. I like it how you also included some made up terms. For completeness, those are not real:
ETag, SAML, CSP, BGP, haproxy, helm file, k8s, SOCKS5, Lamport OTS, Winternitz OTS, SPHINCS, lattice-based, pg-vector, sameSite, httpOnly. (Or, at least, i have no idea what those are ::)
Don't set your goals so low. We already reached 17k on a small models.
Since the whole goal of software architecture schemes it to allow the rest of us non-geniuses to still understand it and modify it, perhaps the same could be true of llms.
Perhaps a million-per-second hypothetical (small) model can be more useful than a state of the art big one.
It's worse than that. One side of the "circle" is 40 billion, the other side is 300. Why not just subtract it, and say 260 billion is going one way.
The real story is that Nvidia is accepting equity in their customers as a payment for their hardware. "What, you don't have cash to buy our chips? That's OK, you can pay by giving us 10% of everything you earn in perpetuity."
This has happened before, let's call it the "selling the goose that lays golden eggs scan." You can buy our machine that converts electricity into cash, but we will only take preorders, after all it is such a good deal. Then, after bulding the machines with the said preorder money, they of course plugged the machines in themselves instead of shipping them, claiming various "delays" in production. Here I'm talking about the bitcoin mining hardware when the said hardware first appeared.
Nvidia is doing similar thing, just instead of doing it 100% themselves, they are 10% in by acquiring the equity in their customers.
I have found it that if you invest some time in learning how to write quality for loops, the quality indeed goes up.
Also, when writing for loops, you have to explicitly think about your exit condition for the loop, and it is visible right there, at the top of the loop, making infinite loops almost impossible.
There is an informational asymmetry, so rogue exchanges can lie all they want without you knowing for sure, and those that do not want to give you your money for whatever reason always do (lie that is).
This happened with MtGox and permanent withdrawal problems, both with USD and BTC.
With USD, it was "Oh, no, its the banks, they are limiting us to $50.000 per day, it's not us".
With BTC, it was "Oh, no, Bitcoin network has a bug, we cannot process withdrawals, it's not us".
LLMs giving you something novel would be like if you let a model play chess against itself and become the best player in the world this way. Totally impossible.
/sacrasm