Northwestern Mathematics races to adapt after wave of AI math breakthroughs

A visual representation of OpenAI's proposed solution to Navier-Stokes, one of the Millennium Prize Problems. Shows many spiraling strands spinning around a central axis supposed to represent a singularity in a vortex.
As artificial intelligence solves many long-standing mathematical problems, Northwestern’s mathematics department is racing to adapt.
WNUR News
WNUR News
Northwestern Mathematics races to adapt after wave of AI math breakthroughs
Loading
/

Over the past two months, artificial intelligence has upended the world of mathematics. After claiming a breakthrough on the Navier-Stokes equations in September, one of the seven Millennium Prize Problems with a million-dollar prize attached, OpenAI released a massive batch of more than 700 mathematics research papers on October 6th. Many of these papers claimed solutions to problems that stumped human mathematicians for decades.

Inside Lunt Hall, Professor Bryna Kra, former president of the American Mathematical Society, says Northwestern’s mathematics department is reckoning with a field that has transformed overnight.

“Math has probably changed more in the last two months than it’s changed in the last 100 years,” Kra said. “One thinks of math as an old, established subject, and it is. But what has been going on right now has been changing all aspects of what we can do, what we can understand, how we do it, what kinds of problems we can approach, and the way we do math. It’s been an explosive eruption in what it means to be a mathematician.”

Even for faculty like Kra who knew a major proof release was coming, the scale of the October 6 release was a shock.

“Many of us were in front of our computers waiting. The extent of it was still surprising when you’re scrolling down and it’s a list of hundreds of files, many of which you recognize what the problem is that it’s talking about,” Kra said. “That’s a little bit breathtaking.”

This rapid acceleration is already colliding with active faculty research at Northwestern. Mathematics professor Ethan Sussman has been collaborating with a mathematician at AI company Anthropic, using large language models, also referred to as LLMs, to tackle part of another Millennium Prize Problem, a famous unsolved physics and math question known as the Yang-Mills existence and mass gap.

“I first started noticing that LLMs were capable of doing graduate-level mathematics back in November or October of last year,” Sussman said. “So about one year ago, and in the space of about one year, it went from being able to do problem sets for a graduate-level course to being able to solve the hardest problems in all of mathematics.”

Sussman’s work involves wrangling tens of thousands of pages of dense, AI-generated calculations, which he calls “slop,” and using language models to distill them into a readable 1,000-page proof. Sussman stresses that the issue with this machine output isn’t its math, but how it is written.

“When I call it slop, I don’t mean that it’s mathematical garbage. I mean, it’s probably correct, but it reads like garbage,” Sussman said.

After the Yang-Mills problem didn’t appear in OpenAI’s October 6 batch, Sussman and his collaborator kept working. On October 25, they posted a breakthrough on the final stepping stone before the main problem, only for OpenAI to announce success on the problem the very next day.

“It happened to be that we posted our stuff online the previous night,” Sussman said. “We didn’t have any insight or forewarning that OpenAI was going to announce the next day. It caught us off guard.”

As of November 7, OpenAI hasn’t released a verified manuscript for their claimed proof yet, and Sussman is continuing his work.

Meanwhile, Kra says some Northwestern graduate students have seen their Ph.D. thesis problems solved by language models mid-degree.

“There are students whose thesis problem, now somebody else has put it into an LLM and posted it somewhere,” Kra said.

Because of those disruptions, Kra says the department is rethinking what a math Ph.D. looks like and which skills matter most.

“Exposition is extraordinarily important now. Being able to explain what is in some of these papers that are honestly a bit incomprehensible to humans at this point, that is a skill that we need to value,” Kra said.

In undergraduate classrooms, foundational courses haven’t changed, since students still need linear algebra, probability, and real analysis to understand how LLMs work. However, because AI can easily solve complex proof assignments, the department is now prioritizing in-person exams.

Senior Brian Kim, a double major in mathematics and applied mathematics, says homework is now graded mostly on completion.

“Northwestern’s math department actually just changed quite a lot. The homework is now just based on completion; it’s only worth 5 percent,” Kim said. “I think it used to be worth much more before GPT.”

Kim says that change has also emptied out departmental office hours.

While Kim says the AI breakthroughs partially influenced his decision to apply to more engineering-focused graduate programs rather than pure math, other undergraduates are weighing mixed emotions. Math major William Wang, a junior, is excited by the new tools, but remains cautious about his future.

“I think in the future we might be able to focus more on asking important questions rather than working on those abstractions,” Wang said. “While I’m also a little bit afraid because I don’t know whether I’ll be good enough in the field before AI just kicks us to the curb.”

Wang says he hasn’t seen peers drop the math major, though he is keeping his post-graduation options open.

“I’m also trying to apply to jobs right now. I’ll still most likely go to a Ph.D., but it’s nice to have some second options,” Wang said.

To help students and researchers adapt, Mathematics Department Chair Tuca Auffinger established dedicated Math + AI committees over the summer, even before the autumn proof releases.

“If you had asked me, hey, when you set up the committee, were you thinking that there would be 700 papers coming out from OpenAI in all possible subfields of mathematics? I would say no,” Auffinger said. “But we saw that there was something big coming to the point that we actually had to take action.”

When the committee hosted its first panel in Lunt Hall to discuss what AI was doing in mathematics, student and faculty turnout far exceeded the room’s capacity. Auffinger estimates there were over one hundred students in attendance.

“I want to emphasize that we started before all the press and the big events. In some way, we’re not reacting—we somehow predicted a little bit about the future, and the system was in place already,” Auffinger said. “Not many universities can say that.”

This fall, taking advantage of Northwestern’s later start, Auffinger, Kra and Director of Graduate Studies Jared Wunsch launched a first-of-its-kind experimental graduate course called “AI Practice for Mathematics.” The course covers the mathematical foundations behind AI models, hands-on lab sessions using LLMs to tackle open math problems, and seminar discussions on ethics. Auffinger says the class has drawn intense interest from colleagues at other universities, and watching students wrestle with AI outputs has been a major success.

“People are learning mathematics and they’re going to the board and explaining a lot of mathematics they got out of the LLMs,” Auffinger said. “So it was not about, “Oh, here’s what the LLM told me.” People were actually spending the time to actually understand it, explain to each other, and that experience for me, it was a huge win.”

The department plans to offer an undergraduate version of the course in Spring 2027, alongside new undergraduate classes on the mathematical foundations of deep learning and matrix analysis. They have also secured funding to give students and faculty access to commercial frontier models.

While faculty still debate how to protect traditional human research, Kra sees the upheaval as a unique opportunity for students entering the field.

“I don’t think anyone can make accurate predictions of where math will be in five years,” Kra said. “But I do know that it’s an extraordinarily exciting time for math because of the range of possibilities that we now have that we didn’t have a couple of months ago.”

For students who want to join the conversation, Auffinger says the department’s upcoming panels are open to the entire campus community.

“So if you’re interested to learn more, please come November 11, 4 p.m., Lunt 105, where there will be an open forum to discuss the abundance of proof in mathematics,” Auffinger said. “It’s just one of many events that the math department is organizing at the intersection of math plus AI, and it’s been absolutely fantastic, a blast. Come early because I’m pretty sure the room will be full.”