AI is dividing the math world in two

Over the last several months, the AI industry has set its sights on some of the problems that have stymied mathematicians for years. In some cases, AI models have made huge progress — though exactly how much isn't always obvious. The Verge's Robert Hart explains what OpenAI and others have actually accomplished, the disastrously bad ways the companies have interacted with researchers, and whether there's a future left for human mathematics. Further reading: ⁠Spotify pulls 50 million AI slop and spam tracks from its recommendations⁠ ⁠Adobe Lightroom mobile can now open closed eyes in photos with a simple tap⁠ ⁠Wii Sports is back on the Switch 2, and it rules⁠ ⁠All the drama around AI’s takeover of mathematics | The Verge⁠ ⁠OpenAI keeps bulldozing mathematicians | The Verge⁠⁠ ⁠OpenAI just wants to win | The Verge⁠⁠ ⁠The AI takeover of mathematics has begun | The Verge⁠⁠ ⁠‘Pure insanity’: Mathematicians will need years to make sense of OpenAI’s latest drop | The Verge⁠⁠ Subscribe to The Verge for unlimited access to theverge.com, subscriber-exclusive newsletters, and our ad-free podcast feed. We love hearing from you! Email your questions and thoughts to [email protected] or call us at 866-VERGE11. Learn more about your ad choices. Visit podcastchoices.com/adchoices

• 0:00Speaker 00

  Support for the show comes from Satellai. If you've ever thought, there has to be a better way to manage my pet's health and safety, then this message is for you. The Satellai Collar Go is an AI-powered GPS dog tracker that helps you keep tabs on your dog, especially in locations where they might run away. And Satellai Sensei can help you spot changes in your pet's health. While others just track location, Satellai decodes lives. Go to satellai.com to get 35% off with promo code VERGECAST. That's S-A-T-T-E-L-A-I.com to get 35% off with promo code VERGECAST.

• 0:40Speaker 00

  Satellai.com, promo code VERGECAST. This episode is brought to you by On Investing, an original podcast from Charles Schwab. Each week, hosts Liz Ann Saunders, Schwab's chief investment strategist, and Colin Martin, head of fixed income research and strategy for the Schwab Center for Financial Research, bring you fresh insights on what's happening in the markets and why, and what the implications might be for your portfolio. Join Colin and Liz Ann as they explore questions like, how do you evaluate corporate bonds that look interesting? And what sectors are on the move right now?

• 1:22Speaker 00

  Download the latest episode and subscribe at schwab.com/oninvesting or wherever you get your podcasts. Support for the show comes from Jira by Atlassian. Your work is scattered. A thread here, a DM there, docs everywhere. And it's not only your team searching for context, your AI agents need it too. Jira taps Atlassian's teamwork graph to bring that context together across Jira, Confluence, Slack, GitHub, and more, and makes it easy to share with agents you already use, like Claude, Cursor, and GitHub Copilot. Assign them work directly or connect your agents through MCP. The result?

• 2:05Speaker 00

  Less token waste, more accuracy, and more time moving forward. Learn more at jira.com. That's J-I-R-A.com.

• 2:20Speaker 01

  Hello and welcome to The Vergecast, the flagship podcast of the Navier-Stokes problem. I'm your friend David Pierce, and today on the show, why is OpenAI suddenly solving all of these big math problems? And what does it mean for the future of being a mathematician? But first, here's everything else happening on The Verge today.

• 2:37Speaker 02

  I'm Josh Kahas, and this is 90 Seconds on The Verge for Tuesday, October 20th, 2026. Spotify announced today that it has removed over 50 million AI slop and spam tracks from user recommendations over the past year, while also changing its system to highlight more human-made music. Other streaming platforms like YouTube Music and Deezer take a tougher stance. They put labels on AI songs and filter out AI slop. Spotify, on the other hand, doesn't mandate AI labels on music. It even partnered with Universal Music Group to launch a tool for creating AI music remixes. So, hooray for humans, I guess? Meanwhile, Adobe announced yesterday that the mobile version of its photo editing app Lightroom can now open closed eyes in photos. The app uses generative AI to match the lighting and pose when generating these replacement

• 3:21Speaker 02

  eyes, as well as reference other images of the same person in your Lightroom library to create an accurate set of peepers. This immediately reminds me of the Wesley Snipes blade meme where the studio added CGI eyes to his face because he refused to open them. And finally, Switch 2 owners rejoice. Nintendo Switch Sports Resort launches today with 12 sports, including two fan favorites from the original Wii Sports, tennis and bowling. The game takes advantage of the Switch 2's motion controls to allow you to jet ski by using the Joy-Cons-like handlebars and use its mouse controls to skateboard. Sports Resort even takes place in an open world, complete with a sightseeing mode where you can explore the island via land, water, and air. Your move, Grand Theft Auto 6. You can read more at theverge.com.

• 4:01Speaker 02

  That's 90 Seconds on The Verge for Tuesday, October 20th, 2026.

• 4:07Speaker 01

  Over the last few months, OpenAI in particular, has started to release what it calls answers to some of the biggest problems in math. Problems that professional mathematicians have been working on, in some cases for decades, even a century. Some of these are big problems that everyone has been working on, that there are big prizes for solving. Some of them are just things that researchers have been noodling on forever. This has given the math community kind of a shock. People are worried about what OpenAI is doing. People are worried about whether OpenAI is taking credit from them. People are worried about what this means for the existence of the life of a mathematician going forward. The Verge's Robert Hart has been covering all of this. He's been talking to professors around the world. He's been dealing with the problems.

• 4:50Speaker 01

  He's been learning what the Navier-Stokes problem actually is. He's going to guide us through all of what's happening and what this means for everyone else who does other things that might someday have AI pointed at them. Let's get into it. Robert Hart, welcome back to the show.

• 5:03Speaker 03

  Yeah, hi. Thanks for having me.

• 5:05Speaker 01

  You are now semi-officially the Verge's expert in frontier mathematics. How does this feel?

• 5:12Speaker 03

  A worrying prospect for everyone involved, I imagine, but exciting.

• 5:17Speaker 01

  Has it actually been, like, for me, one of the fun things about being a journalist is like randomly getting to suddenly get at least a little bit of knowledge on something? Or it's like, there's no shot of me being an expert on this thing, but I now have an excuse to get like weirdly into this esoteric thing for two weeks? Has this been the last like three or four months of your life, just learning math terms?

• 5:41Speaker 03

  Yeah, I mean, I've tried learning them before. I've tried at university. I can't say I did particularly well, far from doing a maths degree as well. But, uh, yeah, I mean, I've covered a lot of science in my past, but mathematics has always kind of been a tough one to get into. So it's, I mean, it's exciting and it's, I think, at heart, a people's story as well. So.

• 6:02Speaker 01

  I agree. So, and this is actually why I think this is so interesting, because I, the, as I first started learning about what's been going on between the AI industry and the mathematics world, it seemed sort of esoteric and niche, right? Like, this is this one group of people who are very affected by this thing, and I want to get into it because I think that group of people is interesting and important. But I actually think there is a much bigger, more broadly applicable thing happening here. But let's start a little bit at the beginning. And I think the easiest place to start is probably with the labs themselves, and I guess specifically OpenAI, which at some point in the relatively recent past, seemed to have turned their attention to solving big math problems. Can you just kind of catch me up on the last few months of activity in this space from

• 6:47Speaker 01

  these labs?

• 6:48Speaker 03

  Yeah, I mean, probably best saying is, oh, the labs have been focused on this for years. Like, mathematics has been a, I think, a big focus of them for a very, very, very long time. And it's kind of, a lot of mathematicians I spoke to say that's like, oh, it's viewed as this kind of, rightly or wrongly, whatever, as a kind of pinnacle of knowledge. And it's, I think, easy for them to try and sort of show off in that area. And, yeah, and it's always been quite terrible. And then the last few months, very good, it seems. And like, and by very good, it's kind of on par with the best human mathematicians.

• 7:21Speaker 01

  So a little bit like the coding phenomenon kind of nine months ago, that all of a sudden it went from a thing computers can do, that it seems like AI might someday be able to do, but isn't quite there yet, to all of a sudden overnight, oh my God, this actually works?

• 7:34Speaker 03

  Yeah, and I mean, well, not that half the time you can even tell it works. They sort of rely on the mathematicians for that. But, um, but, yeah, basically, it's like, there's all these problems that are kind of labeled as open problems in the field, and they've just kind of been mowing through them. And it's gone from kind of problems no one really cared about or noticed, or, oh, I've maybe heard of that here or there, to ones that, oh, some very accomplished mathematicians have spent their entire careers trying and failing to do this, and they did it in like three days. So, yeah, it's a very quick, quick change.

• 8:11Speaker 01

  Yeah. And were, were the people steeped in this on the mathematics side surprised by this? Like, obviously, this has been a surprise to those of us who are not paying close attention. But there's a, there's an industry of people who, you know, spend their days and nights steeped in this stuff at universities all over the world and stuff like that. Were they caught off guard by this sort of turn from it doesn't work to it works?

• 8:35Speaker 03

  Yeah, I mean, I think a lot of that perhaps depends on where in mathematics you look for it. It's kind of a field that varies between, you've got some who are very much like, I use my chalkboard and my chalk, and that is how I work, through to they're using these types of tools, and so they're quite familiar with it. But even among, I'd say, the kind of really early adopters and the enthusiastic users, a lot of them that I spoke to were still quite surprised that we'd got here now. They thought that some of the problems we're seeing going down, maybe give it a few years, possibly, and then within the space of like a couple of weeks, they're like, oh, okay, I'm readjusting that timeline and figuring out what that means now.

• 9:17Speaker 01

  Interesting. Well, and yeah, walk me through the figuring out what it means, because I think, when you say this is a human story, I think this is a lot of what we're talking about, right? There is this, this thing that happens, and I think a lot of people are either dealing with the reality of this already, or the theoretical possibility of this coming soon, where a thing that I have spent my life learning how to do and doing, all of a sudden, there is this massively resourced AI lab over here that has turned its like Eye of Sauron firepower on me specifically, and is going to run me out of a job all at once. What has this felt like for the people at the sort of tip of the spear in the math

• 9:59Speaker 01

  world who are suddenly finding themselves competing with OpenAI to solve these problems?

• 10:03Speaker 03

  Yeah, I mean, the Eye of Sauron, funnily enough, is something that came up, actually, probably around three times with different mathematicians I spoke to, because it's, it's not evenly spread across this field either. So someone like, oh, is it coming my way? Is it coming my way?

• 10:20Speaker 03

  Maybe, who knows. But and that adaptation, I think, this was how people reacting is happening so quickly that people almost don't know how to feel. And I mean, I first started covering this more diligently, maybe two months ago, maybe three at best, when there was some sort of progress with an internal model that looked impressive. And then it escalated massively from there. And ever since, I've had a lot of mathematicians sort of texting me or something. I'm really worried that the number of them that messaged me saying, I'm struggling to sleep, or I don't know whether there's any point in continuing because, I mean, OpenAI were quite clear after one of their big problems they solved that like,

• 11:03Speaker 03

  we've got more than 100 waiting. That was it. So it was that sort of like, is the eye of Sauron on my discipline already? Should I bother? And the reactions really were from kind of despair to, let's just carry on to, I'm going to race to put out as much material as I can possibly put out that I'm working on before I get scooped by OpenAI or someone else.

• 11:30Speaker 01

  Yeah. And it seems like, obviously, there are a lot of people feeling a lot of different things, but it does seem like one of the overwhelming reactions to this has been frustration with OpenAI in particular, not necessarily just for advancing the technology. Like, one obvious thing to feel here is a sort of professional jealousy and fear, right? That like, oh, if the computer can do my job, now what am I here for? Right? And I think, again, that's a thing a lot of people are experiencing right now. And I think that's very real. But it seems like a lot of people you've talked to are able to set that aside and say, even if it's cool that we now have these tools that are able to crack these problems, even if I can partner with this thing to do work, I've never even been

• 12:14Speaker 01

  able to dream of before, and that could be very exciting, still the way OpenAI is doing this feels bad. Can you tease that apart for me? Like, what is it about the way OpenAI in particular is handling this that is rubbing all these mathematicians so much the wrong way?

• 12:29Speaker 03

  Yeah, I mean, it's almost kind of like Silicon Valley comes to insert area here. Like, it's that very sort of move fast, break things and move on before you've even kind of, not even picked up the pieces, but even seen the pieces fall.

• 12:44Speaker 01

  Right.

• 12:45Speaker 03

  And it very much feels like that, I think, with a lot in the area. And like, the conduct is, I mean, it's quite a world away from what happens in a lot of academia anyway. I mean,

• 13:01Speaker 03

  I mean, so scooping happens. Like, that's one thing I think a lot of focus has been on whether or not they've scooped researchers because their Millennium Prize problem, that they get their kind of, I suppose, crowning achievement in this. The big allegation they had here was that they, and this is by their own admission as well, which is almost quite stunning. They'd heard a rumor that some researchers were getting close. And so they then decided to basically throw everything to see if they could. They spent, I don't know quite how much, rumor has it, it's in the tens of millions of dollars, to spin up about 10,000 agents. And in, I think, 80, 88 hours odd, they'd done it.

• 13:40Speaker 01

  Wow.

• 13:40Speaker 03

  And it was a big achievement, but it just feels a bit, it's an unpleasant taste it leaves in the mouth when you sort of, the only reason they did it was someone else was, which again, it happens in academia.

• 13:51Speaker 01

  Right.

• 13:52Speaker 03

  I think what is new is it's quite rare that they can turn their eye, an academic could turn their eye to anything and do it capably. And the resources as well, like, it's without precedent, really, that level of scale. And I think that's what's, that's one element that I think has left a lot of people thinking like, it's against the norm.

• 14:13Speaker 01

  Yeah, I'm sure your average maths professor is sitting here being like, gosh, if you give me tens of millions of dollars and unlimited compute, I could also do some pretty interesting stuff. Like, wouldn't it be great if we were fighting a fair fight here? It feels like a not reasonable place to land for these folks.

• 14:27Speaker 03

  Well, too, I mean, the fairness bit comes to it as well. I mean, OpenAI throughout have said that we want to build tools that let scientists do great things, which is a lovely aim. Why aren't you giving the scientists, or mathematicians in this case, like, the tools? Like, this is an internal model that is not publicly available. No one can even verify. But there's another element as to what they do as well is that, obviously, preprints have been a big thing within the scientific community for a while now, but they're not peer reviewed. They're not necessarily checked. And in a lot of cases, that's happening here, they're written by AI, which we all know is really high quality and really easy to understand, especially when dealing

• 15:10Speaker 03

  with really complicated material. And so what they did most recently is, there are more than 100 results. And it turns out the more than was doing quite a lot of work. There was nearly 400,

• 15:22Speaker 01

  Whoa.

• 15:23Speaker 03

  At which they'd spread out on, I think, more than 700 preprints that they just dropped one evening.

• 15:31Speaker 01

  Wow. And this is, this is like math problems that have been out there unsolved with people working on them for a long time. And OpenAI was just like, we fixed them all.

• 15:39Speaker 03

  Yeah, except

• 15:40Speaker 01

  Good lord.

• 15:41Speaker 03

  Except it was a big question mark there because not all of these have been. So there's a thing in mathematics, you can formalize something, so you can kind of prove it computationally, even if you don't understand it. I think, like, not all of these had that with it. So you can't easily check that it's necessarily right. Obviously, it's specialized material. It's difficult to go through these papers. One mathematician I spoke to was like, if this was written by a human, it would have, I'd be surprised they were able to write it, kind of thing. It was that badly written. Attribution is all over the place, which is a big thing with AI, as we know. And so like, a lot of it, the norm there is that they're not only sort of saying, oh, well, we've solved all these problems. Aren't you grateful? It's dumping a load of work on the field as well.

• 16:26Speaker 01

  Right.

• 16:27Speaker 03

  That fundamentally also misunderstands the field. I mean, I think a lot of people, and I'm myself perhaps guilty of this in the past, are thinking that mathematics is sort of a list of, let's solve this, this, this, and that. Which is part of it, sure. But like, there's a lot of value in the kind of understanding whether it's fit, forging new methods or kind of understanding that. And that all comes when humans solve problems is a natural part of that process. We've now just got that end point, maybe, if it turns out to be true.

• 16:57Speaker 01

  Right.

• 16:58Speaker 03

  And all of that other bit is kind of missed as well. And that's, I think, where it's kind of annoying. Because, I mean, they've said this most recent drop of sort of 400 odd results, I've been told, will take years at least to sort through.

• 17:10Speaker 01

  Wow.

• 17:11Speaker 03

  Just on a basic level.

• 17:14Speaker 00

  Support for the show comes from Shopify. Starting a business requires you to be resourceful. And the more you flex that muscle, the more you start to see opportunities for growth. For example, launching your business with Black Friday approaching means you can take advantage of the extra online traffic. Shopify is there to help you make it all happen. Shopify is the platform for a record breaking, reliable holiday selling season for businesses at every stage. Their pre-built website themes and AI site builder make it easy to launch a beautiful, professional looking store. And you don't ever need to touch a line of code to do it. They also have a mobile app that lets you run your store from anywhere. Their AI e-commerce expert, Sidekick,

• 17:56Speaker 00

  can help you manage almost any business task with a single prompt, including holiday discounts. With Shopify, you can easily sell everywhere people shop: online, in person, across social media, and even in AI chats. If you're ready to launch your business before Black Friday, go to Shopify.com/vergecast to start your free trial. That's Shopify.com/vergecast. Go to Shopify.com/vergecast today to hear your first

• 18:31Speaker 00

  Support for the show comes from Superhuman. There's a lot of potential for AI assistance to make your job easier, but not if every ask starts with you debriefing the entirety of the project from scratch. Ideally, it would be following along from the start and ready to jump in as soon as you need help. Superhuman says they've got the solution. Superhuman Go comes from the makers of Grammarly. It can embed in the tools you and your team already use, always ready to help without having you switch apps or break your flow. It can help make sense of a dense message thread, help you get ready for a meeting, or catch a typo before you hit send. And when you get stuck, Go can help you brainstorm, outline, or review relevant contexts so you don't lose your place. It's basically an AI assistant that's always on, always ready to

• 19:16Speaker 00

  help. Bottom line, Superhuman is built for people who take their work seriously and expect their tools to do the same. You can try Superhuman Go from the makers of Grammarly. Find out more at superhuman.com.

• 19:33Speaker 00

  Support for the show comes from Sonos. Think about the first time you heard your favorite song. Who were you with? What did it make you feel? Was it the music or the lyrics that grabbed you first? And what were the specific tech specs of the sound system it was playing on? Yeah, nobody remembers that last part. Sonos gets that. Their system is designed to just be there, clear, ready, out of the way, so nothing stands between you and the next song that hits you out of left field. For a bunch of years now, I've been incredibly jealous of my friends who have Sonos speakers. I was just always impressed how all that sound and punch and energy came from that speaker in the corner. But Sonos recently sent me two of its Play speakers,

• 20:17Speaker 00

  and now I truly, truly get it. Everyone knows that the quality is immediately noticeable when you listen to a Sonos speaker. But the bigger thing that I've learned is how everything is working together and creating this little ecosystem. It's just really important to me, especially as audio has become more integrated into our daily media lives, whether that's music, a movie, and yes, the Vergecast or Virgin History or Decoder, I want it to sound as good as possible. Visit Sonos.com and build your system. Sonos, love the sound, feel the system.

• 20:55Speaker 01

  This is the part of all of this to me that is, I think, the most sort of generalizable in really fascinating ways and is a thing everyone on both sides of this question ought to be reckoning with in really real ways, which is there's a thread in a lot of your stories that is people being like, again, look, even if they solved this, even if the thing arrived at the right answer, most of the value of doing this research is in figuring out how to arrive at the right answer. Right? We learn, we learn new tools, we learn new systems, we learn new methods, and it is the work that then gets us to the next thing. And this assumes that just knowing the answer to the test is enough. And if you give us the answer to the test, you have given us everything that is required.

• 21:39Speaker 01

  And they keep just being like, no, that's actually not what this is. Because now what's going to happen is there is no next problem for this AI system to be pointed at because we don't know how we got to the answer to this one. And like, are we now betting on the next OpenAI model to be able to creatively imagine new kinds of problems? Which I think is sort of a leap beyond anything that has been able to happen before. But there's this real sense of people being like, we need to understand what works and how and how we got here. And even the people at the labs don't know. Like, they're just like, we put a bunch of compute at this and it arrived at an answer that we think is right. And I just, I don't, I almost don't know how to bring those two sides closer to one another, given the way this technology actually works.

• 22:24Speaker 03

  Yeah, I mean, I think that's almost at the heart of the problem here is that you have one side who are like, haven't we done a great thing? And it is great.

• 22:33Speaker 01

  Right. We answered your question.

• 22:35Speaker 03

  Yeah.

• 22:35Speaker 01

  Yeah.

• 22:36Speaker 03

  But it's almost like, but that wasn't what we were trying to do. Well, that wasn't all of what we were trying to do. And you've kind of like, made the rest of it worse, maybe on the way. We're not, that's something you can't even tell because it's so, like, some of the papers, I mean, three of those papers that they put out, they already retracted due to what they said was a sign error, which, I mean, I'm sure it's a bit more complicated than it sounds. But like, I mean, that's the kind of thing you make in school, isn't it? And you're like, oh, silly me.

• 23:01Speaker 01

  Right.

• 23:01Speaker 03

  Which feels like it could have been, you know, checked before they put it out with hundreds of others. And it's, it just feels almost like a lack of care, I think, is what a lot of mathematicians spoke to me. It feels like they don't really care about the field, they just care about, and I mean, the headline for one of the stories, it was like, OpenAI just wants to win. And I feel that is a reasonable kind of attitude for a lot of how these big AI labs work. They want to get something big and they want to get there first.

• 23:31Speaker 01

  Yeah, I was thinking a lot about the race to AGI in the context of the OpenAI just wants to win thing, that what happens in so many of these cases is these companies are given some kind of brass ring, right? Like, we're going to achieve AGI, we're going to solve a Millennium Prize math problem, we're going to, we're going to do XYZ thing successfully. And all that matters, all that matters, all that matters for your promotion, all that matters for your company valuation, all that matters for your reputation in the tech industry is whether you did it and whether you did it first. The end. And that is just the system we have built. And especially with AI, which is just this giant hammer looking for nails at all times.

• 24:14Speaker 01

  That, it feels like that has become so exacerbated. And the idea of having a thing that builds on itself and that there is a community of people learning together and that all of that is actually designed successfully to make progress happen is just falling apart in this race to just hit milestones because there are things we can measure that look like success. And again, this is, this is maybe just toothpaste that has been out of the tube for some time. But we're in a moment where anything that is quantifiable, everyone in tech is just racing towards as fast as they can and burning the boats behind them. And it's just the culture of that next to, and particularly this academic culture that has been so different and is so deliberative in its way.

• 24:58Speaker 01

  Again, I just, I have a hard time figuring out how those two things come together. But I will say, to, I was going to say to their credit, but even that might be too generous. It does seem like there is some effort on both sides to figure out how this can be slightly more harmonious. What are you hearing? Is any of this working? Is OpenAI learning how to be a better citizen of the math community?

• 25:19Speaker 03

  Well, it's, it's complicated, I think, on that. I mean, it's trying, or at least it's saying it's trying.

• 25:27Speaker 01

  It's not doing nothing.

• 25:28Speaker 03

  Like, that is, which is. It's not doing nothing. It has perhaps at times shot itself in the foot in trying to do so, so.

• 25:35Speaker 01

  Weird, OpenAI never does that in other places.

• 25:37Speaker 03

  Never, never, unprecedented. But, so one big thing that they were sort of saying, well, when they sort of resolved this Millennium Prize problem, they said, well, we've got 100 more. We want to be quite responsible with those.

• 25:51Speaker 01

  Right.

• 25:51Speaker 03

  So we're going to consult some elite mathematicians.

• 25:54Speaker 01

  Which, by the way, it seems had the opposite effect of essentially dangling 100 more anvils over the heads of mathematicians for some unknown amount of time.

• 26:01Speaker 03

  Oh, yeah, absolutely. They, like, that was when I started getting texts and people were like, I can't sleep, this is absurd. Or like, I've published three papers.

• 26:09Speaker 01

  They're like, well, we've dropped this one bomb on you. We have 100 more over here, but don't worry about those yet.

• 26:15Speaker 03

  And it turns out it was so much bigger than it appeared either. I mean, which also raises the question, is there something that prompted them to release that many at once? Who knows? But, but yeah, but so ostensibly, they've been trying. And one of that was to consult a group of mathematicians who OpenAI had initially approached, and this was an elite group, I think, it's about maybe nine, nine people. Among them, sort of, they're all kind of top of the field, basically. So it's not representative of the field as a whole, but it's creme de la creme. And OpenAI had initially approached some of these people to form this board, who had responded with saying, well, we'd rather not be an OpenAI thing, we're going to form our own thing. They'd approached a few more, formed a group.

• 26:58Speaker 03

  Except then OpenAI somewhat sort of quickly hijacked their announcement because, I mean, they've got a bigger footprint. And it kind of made it sound like they were OpenAI's thing. And so a lot of mathematicians I was speaking to around that were like, oh, yeah, it's OpenAI's group. Who knows if they're being paid by them? And, and I chatted to a few people in this group and they're like, categorically, nothing. Like, the only thing we have is like, maybe a bit of advanced access to help them communicate it. And that is, that is it. And they've said they've been talking to other AI companies as well about communicating their results, but OpenAI sort of hijacked that and made them look less than independent, which immediately dented their credibility despite the effort. And then OpenAI also made quite clear that, well, we're not bound by what they say either.

• 27:41Speaker 01

  So this is like the Facebook Oversight Board, which is, it's, it's, it's nice in theory and everyone will, you know, feel bad when they do bad things, but there's nothing actually real or binding here if they don't want it to be.

• 27:53Speaker 03

  Yeah, pretty much. And they've taken a few bits on board. So they obviously dropped that sort of bombshell of 400 odd results. The quality is better. I mean, it's poor and variable, but it is a lot better from what I've been told. They've also, one of their other things was to commit to sort of helping foster that understanding element we spoke about. So whether that's through conferences or talks or workshops or that. And I believe yesterday, this is both OpenAI and Anthropic, said that they'd put 5 million each towards, I think it's a Simons Foundation initiative to do just that. I have seen skepticism over the amount of money, considering

• 28:36Speaker 03

  what they spend on the field and the disruption. But again, it is something, but it's again, like, as to, they've also very clearly said that we are not going to be doing what was, I think, arguably the biggest like recommendation or urging from this group was, please stop testing internal frontier models on everything you can find in our field. Please. And OpenAI quite clearly said then, when they denounced these results, we are trying to develop tools for scientists. And as a result, we will be testing basically everything on these frontier models that we can.

• 29:12Speaker 01

  I have to say, my immediate reaction is that A, of course, OpenAI is going to say no to that. And B, it's kind of an unreasonable request to make. I don't know, I need, I need to think more about this. But I am curious, having, having talked to a bunch of folks on both sides about this, what do you make of that request from the math community of essentially saying, if not stop, at least slow way down?

• 29:41Speaker 03

  I think even it's that level playing field we were talking about, I think, that is, I agree. I think to say, please don't test this thing on that is perhaps a big stretch. I also think it's perhaps not a stretch to say, okay, well, there is a history here. There is a tradition here. There is, there are contributions here that you are not acknowledging.

• 30:03Speaker 01

  Right.

• 30:03Speaker 03

  In the early hours of that, kind of the big drop, some of the texts I were getting saying, it looks like they've just got a paper that someone else did like a month ago. Like, I don't know whether or not that hand out, but like, you don't know because of the way they credit these things. And like, it's well known that they have been quite spotty at that in the past, including in the mathematics drops. Like, they've acknowledged it themselves when I've spoken to them. So I don't think that's necessarily like to say, hey, could you level the playing field a little bit, or at least give us a chance, especially when they're using material that, I mean, it's the same argument that you have in like, a lot of these fields with copyrights. So media or writing or anything, it's you've

• 30:42Speaker 03

  used public works for a private model. And I suppose the equivalent here, if you're thinking of images, now none of you are allowed to paint.

• 30:53Speaker 01

  Right.

• 30:53Speaker 03

  Because we've done all the painting, painting solve.

• 30:57Speaker 01

  Like, we've made the best painting.

• 30:59Speaker 03

  And it's galling enough when it comes down to sort of creative endeavors.

• 31:04Speaker 03

  And whilst it may lock it up professionally, it does not stop people from actually doing that. You could argue that, yes, there are elements of mathematics that it doesn't stop either. But I, it feels a slightly different type of like disruption than it would for those other areas. I also think, given that they then, like, one mathematician says, they need us to celebrate their work, or to validate it, to prove it, to know that it's good, and to use it for whatever hype purposes they want. They need the mathematicians. And I think if they're, I don't think they're even asking by dropping it on them, they're demanding, or presuming even worse. I don't think it's that unreasonable to say, could you at least, even if it's not playing fair, not play in what feels underhanded.

• 31:48Speaker 01

  I buy that. It seems to me that the good news of that line of thinking is that that is a relatively straightforward thing to accomplish, right? That, that kind of interaction and symbiosis and communication would not be that hard to do. The problem is OpenAI has essentially no incentive to do it. And this is what we've seen over and over, right? As this is, and it's not just OpenAI, these companies have discovered that they can essentially get their way by just running at such speed with so many resources that you, like, you know, the, the one you just steamrolled is going to complain about it. What are they going to do? They're flat on the ground. Do the folks that you

• 32:29Speaker 01

  talk to on the math side have any optimism that this can happen? Because it does seem to me, like, if I'm going to be a little Pollyanna-ish about this for a second, there is a world in which something incredible is happening here, right? That, that what we've given is these people who have been struggling with some of these problems, a new set of tools to help them not only solve those problems, but invent and solve the next ones. And that we can speed this stuff up. And that, like, the way that we talk about, you know, drug discovery and, and the way that that stuff can work can be applied to all of these other industries. There's something here that feels like it could work. And, and I, I spend a lot of time vacillating between optimistic that we will get there and completely nihilist about it. What do you hear from the mathematicians right now?

• 33:14Speaker 03

  Yeah. I mean, again, it's, I think it's a very mixed bag. I think the speed makes it very hard, I think, for people to imagine where to go. So like, I mean, the disruption of this many results in one go, like, the amount of people, I didn't really speak to anyone who directly had this, but they said they had colleagues whose research programs have blown up overnight.

• 33:31Speaker 01

  Wow.

• 33:32Speaker 03

  What do you do with that? I mean, and I remember in seeing some of the comments in, from Verge readers on the story I had said, well, there's still clearly a job for, for mathematicians, which, correct. There is a lot to do in understanding, contextualizing, or even using these tools, which I'll return to. But the point is, like, does the profession survive that long, long enough, given the speed of this, to do that? Or, or does it change its value system in crediting more than just the end result in time to recognize the value of people doing all that, that other stuff? And that's, I think, the question here is that, because it's happening so quickly and it's being done with seemingly such little care, that those systems in place say, well, what do you do if you then obliterate an entire generation of postdocs and PhD students because you solve their problems?

• 34:15Speaker 01

  Yeah.

• 34:16Speaker 03

  What happens after that? Like, there's then both a need for them and an absence of trained mathematicians. But, but some people, especially, were also very excited about what, what all this shows. And in saying that, well, one is that a lot of what's happened isn't necessarily that creative, opening up new area kind of stuff. So there's either excitement that maybe AI could do that, or that still leaves lots of avenues for humans. The other is that people, I've, I've, I now have people sort of text me, like, preprints or, or papers they're working on, which I, I, I understand all of.

• 34:51Speaker 01

  I love the idea of you at like, two o'clock in the morning, just getting a text that's like, 8,000 words of equation that is just completely unreadable. And they're like, look what I've discovered. And you're like, this means nothing.

• 35:02Speaker 03

  I mean, with, with the time difference as well, that is about the time it would come through.

• 35:06Speaker 01

  That's amazing.

• 35:07Speaker 03

  And I also didn't know some notation could go via text, but it can. And it's, it's fascinating, though. And it's, so it's hugely accelerating research. People already, I was texting a few people today ahead of recording this, saying, oh, how's it going? And they're kind of like, working on that long process of digesting what's there. But they're already building on it, on, on some of the results. And that's really exciting. And using these tools as well. And so, I mean, there's a fear as to whether it will sort of become a pay for play type thing as well, because mathematics isn't notoriously a very, like, flush with cash academic discipline.

• 35:46Speaker 01

  Sure.

• 35:47Speaker 03

  And so even a relatively modest kind of cost for AI could be too much. But it's, yeah, I mean, but a lot of people quite excited. I mean, some people are also very scared of using these tools for reasons of, I mean, some of the allegations that have gone around with, with a lot of these results, in particular, in the early days of the kind of the, the Millennium Prize problem they solved, was that, are these labs using material from the conversations that researchers are having with their tools?

• 36:17Speaker 01

  Hmm.

• 36:18Speaker 03

  In their models, which later go through. And I always think it's a very difficult one to answer.

• 36:25Speaker 01

  Right. In either direction.

• 36:27Speaker 03

  In either direction.

• 36:27Speaker 01

  Hard to prove, hard to disprove. Yeah.

• 36:29Speaker 03

  And so, OpenAI's initial comments with, with the Millennium Prize was, basically, we can't say no, which obviously does not sound great. They have since come ahead and come out and said, like, we did not. I will say, and I've not dug into this, but it is also very precisely worded language that is used. Because as we know, like, there are a lot of things that these models can't, you can't tell whether they've hoovered it up or not, or in what form, or whether this is synthetic data based on conversations, which is not the same as the conversations. And so a lot of people are now quite concerned about using these tools, as powerful as they are for their research, out of possible

• 37:10Speaker 03

  fear that they would be giving their own work away to some competitor. And it's a difficult one to, to do. And again, it's sort of similar with any other discipline that we've seen.

• 37:23Speaker 01

  Well, that, that's kind of the last thing I'm wondering is like, how long until you start getting texts from every other professor at every one of these universities? Like, I, I've got to imagine that everyone in academia is watching this particular thing happen. And again, looking and saying, okay, at what point is it coming for me? And there's something specific about math that it is pointed at these specific kinds of problems. Like, there's not, there is no equivalent of that that I'm aware of in like, the English department at these universities. But this question of, can it, can these things come and essentially attempt to obviate me by quote unquote, answering all of these questions? It does, i- if I'm OpenAI, like, everything about what we know about these companies suggests that, again, they are giant hammers looking for nails. And they will find them

• 38:08Speaker 01

  everywhere they can. And I have to assume you're going to start getting the numbers of every professor at every college around the world.

• 38:16Speaker 03

  Yeah. I mean, the answer to that question, I think, requires me going back in time. I have already started getting those texts.

• 38:23Speaker 01

  Okay, there we go.

• 38:24Speaker 03

  Any listening also, please get in touch. I mean, I've, I've been following it closely for a long time. But I mean, mathematics is an interesting one, I suppose, in this regard, because it doesn't have a practical element most of the time, in that you don't need to go and sort of verify in the lab. Anthropic have quite famously given Claude a lab somewhere in the Bay Area, I believe.

• 38:44Speaker 01

  Oh, I didn't know that.

• 38:44Speaker 03

  I'm not entirely sure on the specifics with working for it, but it's kind of, the hallmarks there have been questions over whether they announced their first big finding a month or so ago. I'd say the press release left a lot to be desired. They likened it to CRISPR a lot, at the same time, which obviously is a huge revolutionary discovery,

• 39:08Speaker 03

  whilst in the same sentence saying, we don't quite know what this does. So, great. And they'd followed it, but basically saying, well, we're still working on all of that, but we wanted to let people know what Claude could do. Which, if ever there was an IPO coming.

• 39:23Speaker 01

  Right.

• 39:23Speaker 03

  There you go.

• 39:24Speaker 01

  Boy, way to not beat the, this is all just to hype your product allegations.

• 39:29Speaker 03

  But yeah, so it's already starting to happen in other fields. There was also similar allegations to whether or not, and I've not had a chance to dig into this myself, Anthropic deny it, but as to whether or not they'd used research or, or work from other researchers in similar areas. And like, but it's the same type of thing you can see happening. It's let's find things that we can throw computer, dot, dot, dot. So I, I've already been speaking briefly, mind, but kind of people in areas, if you think sort of somewhat similar to mathematics in this world, so areas of economics, for example, or computer science, or like, engineering, or kind of elements of those, like, they're all, I think,

• 40:10Speaker 03

  worried under cells the sense of almost impending doom. I mean, a lot are very excited about these tools, but like, what do you do when a giant turns its eye on you and decides to just walk in your direction? You just kind of hope for the best at that point, but, and figure out what happened after. There's a lot of these efforts now to kind of get more data in, say, material science, or biology, or even automating the robotic sort of parts with labs, so they can generate their data and then do some more and do some more. And I, I mean, I've been around labs that are sort of starting to do this, and it's fascinating, but it's all sort of heading in this direction. And

• 40:53Speaker 03

  I mean, at least for some, there are practical applications. So like, mathematics is kind of a pure research basis. But it will be interesting to see where, where this lands when they do start. Finding something that they could immediately be useful. Like, I found it fascinating that Anthropic were very clear going, Claude found this enzyme system, which if you were to try and patent it, would immediately make it impossible to do so, I believe. I mean, there's probably a nuance there, but like, as far as I'm aware, in the US, an AI cannot hold a patent.

• 41:25Speaker 01

  At least for now.

• 41:26Speaker 03

  At least for now. That is when they turn clause to law.

• 41:30Speaker 01

  Yeah, exactly. Yeah, well, it does seem like, I think, we are going to play out some version of this in discipline after discipline and profession after profession. And the only hope is that each one of them, we get a little more sophisticated in how to actually make this work for everybody. But the way it's gone with math so far, at least, I can't say fills me with hope about where it's gonna go.

• 41:54Speaker 03

  No, it does not inspire confidence, does it?

• 41:58Speaker 01

  No. All right, well, everybody, if you are, if you are watching or listening to this, get in touch with Robert. He doesn't get enough 2:00 AM text messages. He was just telling me he wants more people to blow up his inbox at 2:00 AM. Get in touch. We love hearing from you. Robert, great to see you. Thanks for being here.

• 42:12Speaker 03

  Yeah, thanks for having me.

• 42:15Speaker 04

  Company spending is complex. Managing it should be simple. AMX Corporate connects cards, expenses and payments to streamline the way you work. Built to simplify, designed to scale. This is AMX Corporate. Get started at go.amex slash AMX Corporate. Terms apply.

• 42:38Speaker 05

  Your heart can tell you a lot about your health. Apple Watch Series 12 measures your heart rate every five seconds with the most accurate heart rate sensing and wearable. So your vitals app, now with heart rate variability, can tell you when something is off. And your readiness score can let you know when to rest and when to push. Hear the story in every heartbeat with Apple Watch Series 12. The features described are for wellness purposes only and not for medical use. iPhone 11 or later required. Based on Apple conducted study of heart rate accuracy, August 2026. Visit apple.com slash Apple Watch Series 12.

• 43:10Speaker 06

  The action's on the court. And as the official beer sponsor of the NBA, Michelob Ultra is getting fans closer to the game with a chance to win courtside seats, merch and more. Visit michelobultra.com slash courtside to enter. Michelob Ultra Courtside, 26 to 27. No purchase necessary. Open to US residents 21 plus. Begins on October 1st, 2026 and ends on June 30th, 2027. Multiple entry periods. See official rules at michelobultra.com slash courtside for free entry, entry deadlines, prizes and details.

• 43:41Speaker 01

  All right, before we go, I have one more thing for you. I have a movie recommendation. So obviously, there's a bunch of tech movies out right now. You should absolutely go see, you should see everything. The three hour documentary about Elizabeth Holmes. Everybody says that's incredible. You should go see Musk, the, I think, four hour documentary about Elon Musk. Everybody says that's incredible. You should go see The Social Reckoning, which everybody says is kind of eh, but is probably worth seeing. I feel like there's a really fun, like 11 hour day where you just go watch all three of those things. But in addition to that, I just want to recommend a movie that I watched recently that is about technology and kind of went under the radar. It's called Good Luck, Have Fun, Don't Die. It stars Sam Rockwell, who comes from the future to recruit a bunch of people in a diner

• 44:24Speaker 01

  to try to save the world from AI. It's a super weird movie. I like, I don't pretend that it perfectly holds together as a movie every single second. It tells the story of how all these people came together and how technology invades their lives and changes the way that they work. But in it, there's actually a bunch of really interesting ideas about how we think about AI safety and how we regulate AI. Part of the premise of this movie is that we invented AI and then didn't invent ways to control it for decades. And in the interim phase, the world kind of fell apart. And boy, does that feel relevant to every single thing that is happening right now. So it's, it's full of interesting ideas about how we think about technology and how all these things relate to each other and what it means if we let technology run away from our ability to control it. It's also just a really fun movie.

• 45:09Speaker 01

  Sam Rockwell is fantastic in it. It's bizarre and funny and just a good time. I really highly recommend this movie. I think it went under the radar in part because it's a relatively small movie, but in part because Good Luck, Have Fun, Don't Die is a bad name for a movie. And one of my closest held theories is that nothing can survive a bad title. It's like that, that movie Mike and Nick and Nick and Alice. Also a good movie. Stupid name, nobody watched it. So anyway, Good Luck, Have Fun, Don't Die. Fun tech movie. I recommend it. Go check it out. Let me know what you think. That's it for the show. Thank you to Robert for being here and thank you, as always, for watching and listening. If you have thoughts, questions, feelings, again, reach out to Robert. He wants to hear from everybody in all disciplines about this.

• 45:51Speaker 01

  This is an important story for, for him, for The Verge, for all of us right now. Figuring out what is happening when AI points itself at you and your job and your life is the question of our time in so many ways. And we want to hear all about it. Vergecast at theverge.com is the email. 866 VERGE11 is the hotline. We love hearing from you. And as always, the best thing you can do to support everything we're doing here is to subscribe to The Verge. Theverge.com slash subscribe. It gets you all of our podcasts ad free, including this one. It gets you all of our newsletters. It gets you all of Robert's coverage of math and his attempts to explain weird math concepts to normal people like me. Everything The Verge does. Theverge.com slash subscribe. Thank you in advance. The Vergecast is a Verge production and part of the Vox Media Podcast Network. The show is produced by Josh Kahas, Eric Gomez, Brandon Kiefer and Travis Sarchuk. See you tomorrow. Rock and roll.

• 46:39Speaker 07

  The award-winning series The Diplomat is back for its fourth season, now streaming only on Netflix. Keri Russell stars as Kate Wyler. In a stunning realization, Kate connects the dots that Russia's most powerful weapon is missing. And her husband Hal conspired with President Grace Penn in a dangerous plan to steal it, all behind her back. Kate is left reeling as she begins to question Hal's growing relationship with President Penn. Emmy winner Allison Janney returns with Rufus Sewell and Bradley Whitford. Watch The Diplomat, now playing only on Netflix. We just moved into our new house and I didn't realize just how close we were to the neighbors. That's when my friend told me about blinds.com. Blinds.com isn't a medication, but it can improve how you feel about your home. Customers who shop at blinds.com

• 47:23Speaker 07

  experience excellent customer service from consultation to installation.

• 47:28Speaker 02

  Side effects include increased privacy and better sleep. Talk to your partner today and see if blinds.com is right for you. Visit blinds.com to order your free samples.

• 47:39Speaker 02

  Online deal days are back at Kroger. Don't miss your chance for extra savings when you shop, pick up or delivery. You can save $30 on your first online order of $75 or more with your digital coupon. Plus, enjoy unlimited free delivery by starting a Boost Free trial today. But hurry, these extra savings end October 20th. Restrictions apply. See site for details. Kroger, fresh for everyone.