Live data from Hacker News

Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

chatgpt.com

351–360 of 681 posts

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#351
post #344

Earlier quoted context omitted.

Here the search wouldn't have been chosing the coefficients independently. Note that one intermediate variable is a polynomial in the input variables, and it is used in other polynomials. A search over expressions like the ones in the counterexample would have a much smaller search space.

1) How do you know this structure is the correct one a priori 2) You are starting at 10^500 possibilities. "Much" smaller is not enough, the order of magnitude of the order of magnitude needs to be changed. 3) You still need all of the other assumptions, which were completely unfounded Impossible.

No, I’m not saying you would know in advance that it was possible, but sometimes you visit the crystal cave and the diamond is just sitting there, so why not work out the odds?

I learned from poking around that checking the invertibility of a system in C is a much, much harder problem than I thought. Nonetheless if that were no object, let’s say coefficients from -16 to 15 (5 bits) times eight terms times choosing up to cubes (64) times three equations is searchable, especially since you have only the final combination of coefficients in the determinant. It’s not impossible to generate the equations like this Fizzbuzz style.

Edit: no. 2048 possible monomials, to the 24th power, not times 24. Fine, can’t brute force it.

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#352

I'll have a blog post up tomorrow about it but the Jacobian Conjecture counterexample is a very funny cognitohazard for LLM assistants. It's a paradox for modern LLMs: they have enough math skills such that they can easily compute the Jacobian to formally verify the counterargument, but its own knowledge base is locked prior July 19th 2026 where all it knows is that the Jacobian Conjecture is unsolved and a random ch…

The models are just a reflection of how you prompt them. When I posted the counterexample, in dry language with a dry personality setting, claiming credit for it with search disabled, GPT simply verified that it was correct but noted that it could not ascertain novelty or priority and that external verification was required. "Model reacts in crazy ways when my prompt influences it to" is really, really not interesting.

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#353

This is the second ChatGPT shared conversation I've seen today that is truly fascinating. The first one was someone proving another conjecture false by just repeatedly saying "keep going" to ChatGPT: https://x.com/DmitryRybin1/status/2079904005652893709 What a world we live in.

Without any more context, "keep going" seems to be doing a lot of work. The user is placing a lot of faith in the LLM to not make subtle logic mistakes and to take good approaches to each problem. In my experience, even frontier models (such as Fable) are quite capable of getting confused during even simple technical work I've done in the dev ops world. For example: LLM: This package hasn't made it to production. ME:…

Agreed, often you have to step in and stop it from reasoning itself into dumb directions, but occasionally it goes just like the transcript in question.

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#354
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?

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#355

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?

Is he in the room with us?

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#356

This is the second ChatGPT shared conversation I've seen today that is truly fascinating. The first one was someone proving another conjecture false by just repeatedly saying "keep going" to ChatGPT: https://x.com/DmitryRybin1/status/2079904005652893709 What a world we live in.

> just repeatedly saying "keep going" to ChatGPT For posterity, this indeed works for most problems where an agent might give up. LLMs don't inherently know something is impossible. The phrase I tend to use in my harder prompts to automate this with a sane loop breaker: > **REPEAT THIS PROCESS UNTIL CONVERGENCE AND YOU ARE OUT OF OPTIMIZATION IDEAS.** You have permission to keep iterating.

I have a pi extension that just runs the same prompt in a loop 25 times. I tried giving it a loop breaker but I found that it'd give up too readily. When the task is actually complete each iteration didn't do a lot of work so it was efficient enough. I suppose another way is to call out to a separate context to check if the task is complete?

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#357

Earlier quoted context omitted.

It reinforces how to "learn AI" is to first master the problem domain. I can use AI for coding after decades of coding. I can't use it for theoretical physics because I can't evaluate the responses.

More and more the skill of being able to ask the right question seems critical to me, and I don't know how one can do that without deeper and deeper domain expertise.

Asking the right questions has almost always been one of if not the most important skills. There are often an almost infinite list of possible approaches, but knowing the domain, you can get close to optimum very quickly. If you just rely on the LLM, it will give you the most common bog standard approach, including the usual bugs and usual quirks.

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#358

Math has some of the most insanely dense and impenetrable nomenclature. I can generally keep my head mostly above water or at least near the surface reading from most STEM fields, perhaps leaning on google/wikipedia a bit, but man, mathematics just so quickly decouples from all common tractable understanding it's insane. Sorry it's a bit of an aside, but I imagine many other otherwise "technical" folks feel the same…

This is also true for almost every other field, even within computer science. The only difference is that a lot of people operate at a very surface level without realizing just how much background knowledge they have accumulated. Think about the number of keywords your average SWE is expected to know. It is rather insane. Cache, stack, heap, process, thread, socket, file, tcp, http, tls, websocks, socks, soc2???, dea…

Exactly. Everyone should experience having to teach something they're so used to doing that they don't even consciously think about it anymore. It makes it very visible how much we rely on unconscious models of the systems we deal with and how difficult it can be to convey that model.

It's why less experienced devs are sometimes mystified seeing a seasoned dev, given a vague description of a bug, guess the cause in code they didn't even write.

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#359
post #319

Earlier quoted context omitted.

> Maybe a year - or two model releases - from now, the AI assistant will be undeniably stronger than Tao, and not an equal anymore. we're kind of well past that (in my opinion), if you consider that this is the same ai assistant that can help you with a recipe, diagnose a weird sound in your car, help with biology homework, translate languages, and so on. even in math alone, i think its indisputably already stronger…

To be fair, Tao's specialty is not algebraic geometry.

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.)"

Re: Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample

#360

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?

Is he in the room with us?

He is known to read HN.
Post reply on HN