UC Berkeley Mathematics Professor Stops Grading Homework Amid Rise in AI‑Generated Submissions

Richard Borcherds, a prominent mathematician at the University of California, Berkeley, said he will stop assigning and grading conventional homework in his classes, pointing to the increasing ubiquity of AI tools that can generate solutions indistinguishable from student work.

Known for demanding coursework and careful grading, Borcherds noted that the swift advancement of large‑language models renders it nearly impossible to confirm the genuineness of written answers. "When a student submits a solution that might have been produced by an AI, I have no dependable method to distinguish it," he remarked, stressing that the assessment’s integrity is now at risk.

His move mirrors a wider unease among teachers regarding how effortlessly generative AI can tackle problem sets, essays, and coding tasks. Although numerous schools have tried plagiarism‑detection tools, such software struggles against output from models that can rephrase or devise novel mathematical arguments. Consequently, Borcherds is pivoting to in‑person exams, requiring students to prove their knowledge under supervised settings.

By switching to on‑campus exams, the department aims to safeguard the fundamental learning goals of its math programs while deterring dependence on illicit help. This shift also fits a rising pattern across universities to overhaul curricula, add oral defenses, and stress project‑based assignments that are more difficult for algorithms to complete.

The change prompts debate over homework’s role in teaching. Should AI reliably handle routine problems, instructors might have to redesign tasks to target higher‑order reasoning, teamwork, or real‑world scenarios that demand contextual judgment beyond formulaic solutions. At present, Borcherds’ courses act as a pilot, and the university is tracking results to see if comparable policies could spread to other departments.