Kevin Buzzard
British mathematicianWikipedia
14
MENTIONS
9
EPISODES
9
PODCASTS
Search complete. 14 mentions across 9 episodes found for "Kevin Buzzard".
Sep 30, 2026
Tim ScarfeHOST
so you're a depth first search kind of guy and when you build something successful that involves lots of other people you have to become more like breadth first search And because there are so many things that are coming in every single day.
Tim ScarfeHOST
And I remember there was a bit in Kevin's book where he said at some point you just kicked a load of people off the Slack channel.
Tim ScarfeHOST
And you
speaker_1ADVERTISER
said, look, I just need to
22 MINS LATER
Leo de MouraGUEST
I can see in the near future, AI is too expensive, right? People would say, well, I'm updating the spec.
Leo de MouraGUEST
Now I have to waste zillions of tokens updating everything,
Tim ScarfeHOST
right? It is really exciting though, because if I remember correctly, Kevin said in his book that I think Z3, that actually solved an incredible number of bugs in Windows 7 before it even got released.
Tim ScarfeHOST
So it was life-saving.
Aaron StumpHOST
And so he had proved this.
Aaron StumpHOST
And I read that Kevin Buzzard, who is a mathematician who's rather famous in theorem-proving circles, for having gotten interested in theorem-proving and also beating theorem-proving people around the head a little bit for not being sufficiently attuned to the needs of math.
Aaron StumpHOST
But this was sort of about pre-lean theorem-proving.
Aaron StumpHOST
So lean...
6 MINS LATER
Aaron StumpHOST
But it was not, like, I mean, these things go relatively quickly.
Aaron StumpHOST
So this, it didn't take, it didn't take human years of human effort.
Aaron StumpHOST
So, I mentioned Kevin Buzzard.
Aaron StumpHOST
Apparently, he had had a project where he wanted to tackle, he wanted to muster a team to tackle a formalized proof of Fermat's Last Theorem.
speaker_0HOST
The proof was filed publicly, in the open.
speaker_0HOST
Kevin Buzzard, who was leading the community effort on that same theorem, compiled the code himself, and signed off on it.
speaker_0HOST
The first attempt had failed.
speaker_0HOST
It only worked with a tool built somewhere else.
Tudor AchimGUEST
And I think that, uh, if you have a process like that and it becomes accepted by the field, then, you know, people that are like, "Oh, well, it can only do the IMO," or like, "Oh, it can only do Airdish problems," or, "Oh, it can only do the double, you know, cyclic cover conjecture," um, it's a little harder to, to be credible when saying that if the entire field has agreed, like, okay, you know, these are the certain steps where we think it's impressive.
Thorsten AltenkirchHOST
I mean, one goal, uh, like Kevin Buzzard is working on is, is formalizing the proof, Fermat's last problem.
Thorsten AltenkirchHOST
Uh, what, what, what do you think, uh, about this? How, how soon will this happen? I mean, I'm, I'm, I'm sure AI would help very much to, to speed up this project, right?
Tudor AchimGUEST
I think it should probably help to speed it up.
CassidyHOST
The headline is, Claude did Fermat.
CassidyHOST
Kevin Buzzard, who actually compiled the code, says it adds nothing to mathematics.
CassidyHOST
And he's right.
CassidyHOST
The feat is dozens of agents producing 13 million lines of lean that a checker can verify.
![[YOU] on AI · Ahora](https://particle.news/cdn-cgi/image/format=auto,width=128/https://cdn.particle.pro/url/media/4e91d5b0-2117-58a6-bbc5-99f50507c704/4ba59d61cb1532c3192cbd071151c34c20f282a4cc8dbcff1e4711bdb0859bab)
AmberCORRESPONDENT
El resultado abarca trece millones de líneas de código y demuestra veintinueve mil quinientos teoremas intermedios.
AmberCORRESPONDENT
Kevin Buzzard, del Imperial College, quien lidera un esfuerzo de formalización paralelo, compiló el código y lo calificó como un logro extraordinario.
AmberCORRESPONDENT
Aunque los analistas señalan que esto convierte una demostración ya conocida en una forma verificable por máquina, no se trata de matemáticas nuevas.
Edo SegalHOST
Para el debate sobre la AGI que gira en torno a GPT-6 Astra con Daniel.
speaker_5HOST
Yes.
speaker_5HOST
The final result was then fully reviewed and mathematically verified by Kevin Buzzard, a leading mathematician at Imperial College London.
speaker_4HOST
But analyzing this mechanical process leads me to a foundational hypothesis.
Etienne NewmanADVERTISER
Okay.
JamieHOST
Anthropic says Claude autonomously formalized the complete proof of Fermat's last theorem in the lean proof language, with Claude agents running for 11 days inside a multi-agent harness called Prove2Me to produce roughly 13 million lines of lean-for and prove over 30,300 intermediate theorems.
JamieHOST
lean's type checker verified the result using only its three standard axioms genuine mechanically confirmed work and mathematician kevin buzzard who leads his own funded formalization project on the same theorem called it extraordinary but anthropic's own repository credits 106 upstream files to buzzard's project and to mathlib and claude's first attempt reportedly failed before the third party prove to me tool got added details that cut against the autonomous framing without touching the type checking The lesson travels past mathematics.
JamieHOST
Verifier-gated tasks, where a compiler gives an unambiguous pass or fail, are where massive parallel agent swarms currently deliver their most defensible wins, a narrower claim than the open-ended reasoning story labs usually lead with.
JamieHOST
In the harness, tools, and orchestration world, Google's always-on personal agent, Gemini Spark, can now edit photos, build and share albums, and run recurring workflows against a user's Google Photos library from a single prompt.