Mathematicians feared AI would write a thousand-page proof; it wrote 15 pages
Humans disproved a conjecture about random graphs in 250 pages, built on another 200; AI proved the weaker result inside group theory in a dozen-odd pages. AI's math proofs are generally short and elegant — long proofs are something only humans write.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The first thing models replace is literature search
Erdős's problems have been compiled into a website, and problems marked open aren't necessarily unsolved — the mathematical literature itself is hard to search. One guest was unsure about one of them, threw it into GPT-5, and five minutes later the model produced a reference, effectively telling him the problem was within reach and this was how to do it. He and a few friends had already spent hours on it without being sure they could solve it. They then found a dozen more cases of the same kind. The model's first beachhead isn't proving, but the retrieval and detail execution that consumes the most of a mathematician's time.
The model isn't smarter, it's more willing to bet
A human mathematician's daily life is "gambling with problems": you have an idea, think it might work, try it for hours, days, weeks, then give up; a year or two later you learn someone else made the approach you'd considered work. The model's difference isn't inspiration — it's that it doesn't calculate risk-reward. If a human tells it to do it, it does it to the end. And it isn't brute-forcing: it tries many different paths, very persistently, but can only try a finite number of ideas, pruning the search tree with existing knowledge plus mathematical judgment. In the unit distance problem, the approach may have been thought of long ago; the real difficulty is aligning all the extremely tedious execution details, and that's the part humans easily get lost in and the model almost never gets wrong.
Humans get polluted by their own ideas, models don't
Once a human has gone down a wrong path for a while, the earliest idea gets bound up with everything that failed afterward — in the guest's words, the context window is polluted, and you can't clone a clean session from last week's self and say, don't do this, rebuild your intuition in a different direction; the model can, or at worst it just restarts. Its way of updating "how likely is this path to work" is also more computational: humans automatically lower their confidence in a method after the first failure, while the model seems better at repeatedly adjusting probabilities rather than killing a path outright. So it does backtrack and make mistakes — it just doesn't get dragged down by them.
The model turned a numerical guess into an equals sign
The sphere packing problem has exact answers only in dimensions 1, 2, 3, 8, and 24; the 3-dimensional proof is at least several hundred pages and notoriously ugly. The long-standing best upper bound came from a two-page 1970s paper by two Russian mathematicians, with a constant of 2 to the -0.599d — the solution to some extremely ugly optimization problem. Viazovska's linear programming framework hits the optimum exactly only in dimensions 8 and 24, which is a major reason she won the Fields Medal. What the model did wasn't to compute the constant a bit smaller, but to prove that the framework's optimal value in large dimensions is itself that beautiful asymptotic: an equals sign, not an inequality. One guest spent about six months on this problem during his PhD with zero progress, and finally saw the model complete it with a few pages of complex analysis.
One "push it further" buys a conjectured value
Of the ten problems, only the pair of spherical codes and binary codes interacted: they first had the model use representation theory to improve the code bound, and the model stopped when it was done; so they asked whether it could push further, and the model produced a much more complex representation-theoretic version, and taking that path to its limit actually recovered the conjectured value for sphere packing in large dimensions — the two pieces of work weren't a coincidence but two ends of the same method. This also prompted the host's question: is this the model lacking judgment, or a harness problem? The guest's answer is task orientation: it finishes what you assign and stops, not that it's incapable; ask again and it goes further.
It's not that it can't push, it's that it wasn't asked to
A more honest summary is: this isn't a capability problem, it just didn't want to at the time. Task orientation is a double-edged sword — the model does what you assign, then stops; sometimes it clearly realizes it has made a breakthrough but won't push it to the limit on its own, because that's not what you asked. This leads to the discussion of taste: someone gives a utilitarian definition — being able to solve problems faster with better judgment is taste; others think taste is knowing which problems and which methods are worth investing in. A compromise is to split it: let one model handle taste, and another act as its subordinate doing the long-horizon grunt work; the two don't have to share context, so they don't pollute each other.
AI does in a dozen pages what humans do in hundreds
A sofic group, roughly speaking, is a group that can be approximated by finite groups; previously no one knew whether non-sofic groups exist. A stronger conjecture (Aldous–Lyons) was just disproved two years ago: that work was called a tour de force, 250 pages, built on another 200, drawing on quantum complexity theory, and few people can read it. But proving the weaker conclusion that non-sofic groups exist took the model only about 15 pages, staying inside group theory the whole time and borrowing only some existing group-theoretic results. The guest said that a year ago he would have thought AI's proof would be a thousand pages of incomprehensible stuff, but the opposite happened — now only humans write two-hundred-page proofs.
Proof is no longer the bottleneck; understanding is
The old logic was: proving a result is so hard that the subsequent understanding, absorption, and explanation come along for free — what you proved yourself, you naturally understand deeply and naturally take responsibility for explaining to others. Now the proof bottleneck has been largely removed, and understanding, putting results into the right framework, and explaining them to others have instead become explicit, rewarded contributions. But the ceiling remains: the upper limit of difficulty for mathematical problems is very high, and even if models keep getting exponentially stronger, problems like P versus NP may never be solved. So mathematics is more likely to organize around these big puzzles, with small puzzles handed to machines — and by the way, models both produce exponentially more mathematics and make absorption faster; they're solving the problems they create.
In their own words · checked verbatim
It's not really trying everything. It tries a lot of different things. It's extremely dogged.
It's sort of like your context window is a little polluted, and you can't just make another clone of yourself from last week and say, don't do this, try something else, build your intuition in another direction.
I had actually thought about this problem for about six months at some point when I was a graduate student. And, yeah, I just, I remember making, like, absolutely zero progress on it.
So it wasn't like a capabilities issue. It just didn't feel like it at the time.
It was like 250 pages building on another 200 pages. It uses like quantum complexity theory.
It's like 15 pages maybe. And it doesn't have any of this like very complicated connection with quantum complexity.
Only humans can generate 200-page proofs right now.
I mean, a nice thing about math is that the ceiling for difficulty of a math problem is pretty high.
Figures
| Time for GPT-5 to find a reference | 5 minutes | 4:08 |
| Similar successful literature-search cases | 10 more found | 4:08 |
| Number of open problems in Astra | 10 | 15:18 |
| Length of the shortest 3-dimensional sphere packing proof | several hundred pages | 17:22 |
| Number of dimensions with exact sphere packing answers | 5 (dimensions 1, 2, 3, 8, 24) | 18:23 |
| Constant of the long-standing best upper bound for sphere packing | 2 to the -0.599d | 20:27 |
| Asymptotic constant of the linear programming bound given by the model | about 2 to the -0.601…d | 22:27 |
| Length of the human disproof of the Aldous–Lyons conjecture | 250 pages, built on another 200 | 50:57 |
| Length of the AI proof that non-sofic groups exist | about 15 pages | 51:57 |
Glossary
- sphere packing
- Finding the densest arrangement of spheres in d-dimensional space; a classic geometry problem.
- linear programming bound
- Constructing a function satisfying two properties to turn sphere packing density into an optimizable upper bound.
- spherical code
- Moving sphere packing onto the sphere; equivalent to studying the rate limit of error-correcting codes.
- sofic group
- A group that can be approximated by finite groups in some sense; previously it was unknown whether non-sofic groups exist.
- Aldous–Lyons conjecture
- Any infinite random graph satisfying unimodularity can be approximated by large finite graphs.
- harness
- The prompts, tools, and training pipeline designed by humans outside the model, which together determine final performance.
How to listen
Readers building AI applications or investing, especially those trying to judge whether models can only brute-force search; and researchers concerned with how the scientific workflow will be rewritten.
From [43:45], the introductory groundwork on "what is a group" can be fast-forwarded.