Richard Borcherds, the UC Berkeley mathematician who won the Fields Medal in 1998, has stopped grading his students' homework for correctness.

According to The California Post, he made the change after realizing that student use of AI is now widespread, so paying graders to evaluate answers that may be machine-generated is 'a waste of time and money.'

He now awards credit primarily for whether students complete assignments, and bases grades mainly on in-person exams where electronic devices are banned. He admits he does not know of a better solution and calls his completion-based system 'not a very good system.'

Borcherds told The California Post that grading for correctness 'just adds a huge incentive for them to use AI for homework instead of doing it themselves, and penalizes students for honesty.'

The Daily Californian reports he saw so many students using AI to consistently receive top scores that AI use felt unavoidable.

Detection is no longer practical either: he said there is 'no easy way to tell the difference between human and AI generated work,' because AI rarely makes the characteristic errors it once did.

Online exams, meanwhile, have become 'a waste of time' — he cannot tell who is using AI, but has noticed average scores rise substantially on them.

According to the Daily Californian, UC Berkeley has explored AI-detection software including Turnitin and Rumi, but determined the tools were not reliable enough to adopt as a standardized campus-wide solution.

The Berkeley Division of the Academic Senate instead asks instructors to include statements in their syllabi outlining course expectations and rules governing AI tools, and policies vary dramatically with course level and focus area, per Anne Marie Richard, the university's assistant vice provost for undergraduate education.

A growing database of sample AI-use statements is available in the campus AI Hub, leaving individual instructors in control of their own classrooms.

The grading change is only the visible edge of a larger worry. In an interview with the Daily Californian, Borcherds said that 'if AI continues to improve at present rate, then, in a few years, there'll be nothing human mathematicians can do that AI can't do better.'

His remarks followed OpenAI's announcement that AI agents apparently solved the Navier-Stokes existence and smoothness problem, one of mathematics' most difficult unsolved problems.

Just five years ago, mathematics looked like one of the safest careers from the AI boom; now Borcherds says he has no idea what to advise aspiring mathematicians — even asking ChatGPT, which reportedly said it had no idea what was going to happen.

Borcherds expects mathematics departments to keep operating much as they do now for several years, describing plenty of 'inertia' in the academic system — he compares it to physical paper journals that keep publishing despite, in his view, having lost much of their purpose decades ago.

He notes past technical disruptions displaced other trades, from hand weaving felled by looms to horses displaced by motor cars, but says 'this is the first time anything like this has happened in mathematics.'

His perspective on scale is measured: he told The California Post that 'if the worst effect of AI on society is that it is harder to assess homework then I will be very relieved.'