Berkeley Professor Stops Grading Homework — And the Tech Community Shrugs

When Large Language Models Render the Problem Set Obsolete

A UC Berkeley mathematics professor has publicly concluded that grading homework no longer serves any meaningful purpose, since students can trivially outsource assignments to large language models like GPT-class systems and specialized math solvers. The reasoning is straightforward: if an assessment can be completed in seconds by a chatbot, its score reflects neither comprehension nor effort. This is happening now because the 2025–2026 academic year marks the point where AI assistance has become ubiquitous, frictionless, and effectively undetectable at the undergraduate level, especially in symbolic domains like mathematics where modern models excel. The technological background is critical here: unlike essay-writing, mathematical reasoning was long considered somewhat AI-resistant, but advances in chain-of-thought reasoning and tool-augmented models have erased that assumption. The professor's response — shifting assessment weight toward proctored, pen-and-paper exams — is less a radical reinvention than a return to pre-digital evaluation norms.