Frontier AI programs can now do math better than graduate students.
Those students, it turns out, aren’t taking the news well.
In a paper published this week, researchers at MIT found that a vast majority of grad students and postdoctoral scholars they polled see their career prospects rapidly dimming as AI remakes the field.
Nearly 76% said they are increasingly unlikely to stay in academic math after earning their degrees. 76% said they feel less motivated to do research “as a result of AI developments.”
And in a free-response portion of the survey, grad students laid bare their raw, pent-up feelings about the subject.
Perhaps most striking is a red-hot frustration with academic leadership, which respondents described as alternatively wayward, useless, meek, silent, and ineffectual.
“It’d be great if institutes gave guidance instead of platitudes on this subject,” one respondent wrote.
“We are lacking high-level leadership from across the academic community,” another said.
For the survey, conducted by three MIT-affiliated math students, the researchers polled 268 early-career mathematicians via a Google Form. Questions were multiple-choice and free response. They all centered on AI’s ever-expanding ability to do work they’ve spent years studying.
Notably, the response window closed just after OpenAI released hundreds of solutions to long-standing problems on October 6. The authors caveated the survey, saying it represents the vibe at “only a small window of time.”
“The past couple of months have been exhausting,” the researchers wrote. “We’re being forced to reckon with existential questions about a field we love, all while incredibly uncertain about our own futures.”
Since OpenAI first really broke onto the mathematics scene in September, when it published solutions to the Navier-Stokes Millennium Prize question, the AI-generated solutions have only kept coming.
Mathematicians have described going through stages of disbelief, shock, denial, anger, and, for some, acceptance.
Though some high-level mathematical councils have convened to issue edicts on how AI should be used, it’s not like OpenAI has to listen to them.
OpenAI and Anthropic have, however, made some concessions to the math community. Both AI giants pledged a combined $15 million to a foundation to support mathematicians, and OpenAI established an “advisory group” to judge its math operation.
If you ask students, though, the concessions haven’t changed much for them.
“It is difficult to do research and go to seminars when it feels like none of the individual work will matter much in the future,” one respondent wrote.
