Richard Borcherds won the Fields Medal, math’s top prize. He’s stopped grading his UC Berkeley students’ homework for correctness, and his reason is simple. So many students were using AI to get consistently top scores that he saw “no point” in paying readers to check their answers. Hacker News surfaced the story, and it’s drawing heavy discussion. That makes sense. It’s one of the clearest signs yet that AI is changing how universities measure learning.
Borcherds now gives homework credit for completion: you hand it in, you get the points. He admits it’s “not a very good system.” He also says he has no better option right now.
🧮 What’s Actually Changing
The timing matters. Borcherds spoke right after OpenAI announced that its AI agents had apparently solved the Navier-Stokes existence and smoothness problem, one of the hardest open problems in mathematics. The math community will take time to verify that claim. Either way, Borcherds is already thinking about where this is heading.
“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,” he said.
That’s not a tech CEO pitching a product. It’s a working mathematician at the top of his field, and he has nothing to gain from hype. It also cuts against what many people believed five years ago, that pure math was one of the careers safest from AI.
🏛️ Institutions Are Lagging Behind
Berkeley’s official response shows how hard this is for universities to handle:
- Syllabus statements, not rules. The Academic Senate asks faculty to describe their AI expectations in each syllabus. Actual policy is left to each instructor.
- Detection didn’t work. The campus tried AI detectors including Turnitin and Rumi. It decided none of them were good enough to adopt campus-wide.
- Templates over strategy. Faculty can pick sample AI statements from a growing database in the campus AI Hub.
Borcherds expects a lot of “inertia.” He wouldn’t be surprised if his department looked much the same in a couple of years. His example is printed academic journals, which still come out even though they’ve been “utterly pointless” for two decades.
He’s probably right about the short term. But inertia doesn’t stop disruption. It just means the shift happens unevenly, with some departments adapting fast and others pretending nothing has changed.
🔮 The Next 1-3 Years
Here’s what I expect, based on the dynamics Borcherds describes:
- Homework stops counting. Take-home work turns into practice, and grades move to in-person exams, oral defenses and supervised problem-solving. Completion-based credit like Borcherds’ will spread because nobody has a better option yet.
- Credentials lose meaning. If a top transcript can be produced with AI, employers will put more weight on live technical interviews and portfolio work. That’s already happening in software hiring.
- Career advice gets harder. Borcherds says he doesn’t know whether to steer students toward math. “I even asked ChatGPT this exact question, and even ChatGPT said it didn’t have any idea what was going to happen,” he said. Expect more students to pair math with fields where judgment, context and accountability still need a human.
- Detection gets dropped. Berkeley’s experience suggests that schools trying to catch AI use will lose. Schools that change how they assess students will do better.
💡 What Practitioners and Businesses Should Do
This story is about a math class, but the lesson applies anywhere output used to prove skill:
- Audit your proxies. If your hiring tests, certifications or performance reviews depend on work done unsupervised, assume AI can do that work now.
- Test reasoning, not answers. Ask people to explain, defend and adapt their solutions in real time.
- Skip the detectors. If a major research university couldn’t make them work, your HR team probably can’t either.
- Plan for capability jumps. If frontier models really are cracking famous open problems, the tasks you consider “safe” from automation need another look.
Borcherds compared this moment to looms replacing hand weavers and cars replacing horses, then added that “this is the first time anything like this has happened in mathematics.” The schools that do best will stop asking how to catch AI use and start asking what’s still worth measuring. You can find more details in the original story on Hacker News.