Today two different parties claimed to have made a breakthrough in mathematics. You’re going to hear more about this soon, I’m sure. I’m not a mathematician or a reporter— but I am an academic, and before this either gets buried in the news cycle or drowned in marketing language I want to talk from inside what’s left of the crumbling ivory tower about why we have the norms we do and why it was wrong that they were breached here.

The big announcement was a breakthrough in a problem involving the Navier-Stokes equations— important fluid dynamics math. While the details are complicated and can be better explained by the appropriate experts, for some context, with my humble undergrad physics degree, this is the example I would choose to give if out of the blue you asked me for something difficult. We got two statements. The first was released by NYU math professor Tristan Buckmaster (you can find it here, and you should read it, since I’m not really commenting on the worst of it without firsthand knowledge, and this is quite bad). Long story short, Buckmaster and an Anthropic employee named Levent Alpöge had been working for some time on a particular part of a difficult problem in math, based on the earlier 2023 work of Luis Martínez-Zoroa, whom Buckmaster says “deserves a Fields Medal” for making this work possible.

This kind of statement is more than gratitude to Martínez-Zoroa— it’s an expression of gratitude that keeps knowledge production going altogether. Academia can be tiring and thankless, especially for the majority of faculty, who don’t and will never have tenure, research time and budgets, or security. Credit is the only thing that tight-fisted state legislatures, a censorious federal government, and economic decline can’t take from you.

Much more prosaically, the credit economy keeps academia sharing knowledge. If scholars didn’t have to credit each other, theft would be rampant— every good idea would be something to guard until you have it incontrovertibly in print to prove you got there first. This would make everyone’s work worse: paranoid conferences, few workshops, as little feedback as you can get by with.

That’s why the second announcement was so jarring to see. OpenAI announced today that it had achieved a similar breakthrough (actually a larger one, since it’s on a more general version of the problem, as I understand it). By their own account they heard “rumors” last week that this problem was being solved by Buckmaster and Alpöge, and decided to throw a massive amount of resources at the problem: 10,000 agents working in parallel, outputting hundreds of billions of tokens. If you or I sought to do the same work (assuming you could— they used an internal-only model) with GPT 6 Astra, this would cost you about $3 million, by my estimate (they say the agents used 300 billion tokens, and Astra’s API rate is $10 per million tokens).

Buckmaster notes that the solution OpenAI found happened to be the same route as theirs— is it possible that the company used logs from Buckmaster and Alpöge’s work to steal their work and beat them to the finish line? Alarmingly, OpenAI itself states “we cannot rule out that de-identified data derived from their usage of our products helped improve our models.” But we don’t need to get conspiratorial here, because again, OpenAI admits to hearing about progress and trying to race to it.

In some spheres of life this kind of behavior, while unpleasant, is tolerated, even routine. In academia it can’t be: again, that’s how the whole system works. And especially with LLMs achieving breakthroughs in math seemingly daily, credit for choosing an effective path is one of the only things left to humans in this field. What a cruelty to take it.

The New York Times announcement makes no mention of Buckmaster, Alpöge, or Martínez-Zoroa. OpenAI’s announcement defensively mentions the former only.

I’m not a mathematician, and my work isn’t currently the kind of thing any AI firm can make an exciting marketing announcement about. “OpenAI announces breakthrough in understanding 18th-century conceptions of cause and effect by relating artisan activity to high intellectual publishing” (roughly, part of my current project) would not impress investors, and right now AI isn’t great at long-form writing. But if someone took my social media posts about my upcoming work and threw millions of dollars at solving my ideas before I could, I’d be heartbroken.

The problem here though is not a technology but a series of institutions and people. Nothing any human being on Earth can do could change the fundamental physical truth that you can make sand do linear algebra to do complicated reasoning tasks. But who gets to decide which tasks, how credit for them is apportioned, who benefits? All of this is still up for grabs. OpenAI has proven itself especially mercenary in recent months, and more broadly the profit incentives of corporations as an institution— a product of law, but also culture, history— point toward doing this kind of thing.

Technology possibility isn’t something human beings get to decide, but implementation is. We have made choices about what kind of institutions and norms we have, and we have to make new choices now if we want to preserve and build on the best of what we already have. Academia can be mean, unjust, austere, careless toward the basic needs and desires of its own community. But it can also be generous in a way few other institutions of this size and significance are. I don’t want to lose that. I want to build on it.

We need more than the old academic credit model because there’s a deeper credit problem here too. It goes well beyond either the “scooping” that happened in this case or the “plagiarism” models are often accused of. OpenAI was only able to achieve this at all because of the hard work of, quite literally, all of us. These models work not by stealing particular ideas already embedded in particular texts, but by compressing all of the knowledge and reasoning we have put in our texts— the little model of the world we have built in language. A lot of video games end the credits with a thanks to you, the player— you’re a participant too! Some similar principle should apply here: you (and yes I mean you, specifically, simply as a member of the human community) deserve a little credit and thanks for this. This is one reason academia shares knowledge with the public. Countless people have made my work possible, and I owe my research to them.

There’s no going back, but we already have the values we need to go forward in a better direction— if take the challenge of this moment seriously. Whether we thrive amidst these technological advances will come down to our ability to be generous with each other. Can we decide we all owe each other safe and secure lives independent of whether someone wants to buy our labor? Can we decide to give that to each other? I hope so. Today it looks like some of us failed to even give proper credit for a single esoteric problem to a few mathematicians. It would have cost OpenAI little to more earnestly collaborate or simply wait. What might they do— what have they done— when faced with more difficult and higher stakes questions? A firm like this can’t be trusted with this kind of power.