A mathematics professor at the California Institute of Technology has stopped grading homework for accuracy because AI homework has made it impossible to tell whose work he is marking. Professor Yi Ni, who has taught at Caltech for 17 years, now awards credit based only on whether the work is complete. "Students are using AI to produce answers on typical problem sets," Ni told Business Insider. The change came after he watched chatbots answer problems that once took his students hours, and it has left one of the world's leading science universities rethinking what homework is for. The reporter who sat in on the class, Truman Dickerson, described a room where students followed the chalk work on tablets while the professor worked through problem after problem by hand.
How AI homework changed one maths class
Ni filled eight blackboards with handwritten equations during an introductory mathematics lecture on Friday, with no laptop or projector in sight. The teaching was entirely traditional, but the grading behind it has changed. The AI homework problem starts with the tools themselves. Chatbots that once struggled with simple tasks now solve advanced problems at a graduate level, and nearly every homework task in college mathematics can now be handled by AI. Ni said he cannot tell whether a submitted problem set was produced with AI assistance, so grading for accuracy has become a futile exercise. He now gives credit for completion instead. AI homework now arrives finished, and the professor has no reliable way to check who did the work.
Sathvik Redrouthu, a sophomore mathematics major who completed the course last year, said he avoided AI on assignments because he wanted to understand the material. He said using AI homework to bypass the learning process is like "shooting yourself in the foot." Students who are not passionate about mathematics might be tempted to use AI to get through the requirements, he said, but those who rely on it may struggle when they face in-person examinations where AI help is unavailable. "I personally think that if you leave AI out, then it pushes the average student to learn properly," Redrouthu said. His view points to the gap between finishing homework and learning the subject.
Why AI homework is hard to detect
Mathematics poses a sharper problem than essay subjects. In the humanities, AI-generated text often shows patterns in sentence structure and argument that a reader can spot. Mathematical solutions produced by AI are far harder to separate from human work, because a correct derivation looks the same whoever produced it. AI homework is hard to spot for that reason, and detection tools offer no reliable answer. Ni said the issue runs deeper than cheating. He described a problem in the learning process itself: if the work can be produced without doing the thinking, the practice that builds understanding disappears with it. A wrong answer can also look polished, which makes the checking harder still.
Caltech has no single rule telling mathematics staff how to adapt. Its teaching resources, published through the Center for Teaching, Learning and Outreach, leave instructors to set course-specific expectations. AI homework policies now vary by course, and some syllabuses now state plainly that AI is allowed. The university's own guidance warns that generative AI output can be inaccurate, incomplete or fabricated, so even a correct-looking derivation needs checking by someone who understands the mathematics. The course Ni teaches is a core requirement, so every Caltech student must complete it. That reach is part of why the change has drawn attention beyond the campus.
What AI homework means for exams
The shift leaves examinations carrying more of the weight. In-person tests still require students to work without AI, and they remain the place where understanding is tested directly. Ni has not said he will change the exams, and the completion-based homework grade still rewards students who do the work. The open question is whether students will keep doing it when the marks no longer depend on getting the answers right. Other universities face the same choice, and the answers they reach will shape how much homework counts in the years ahead. For now, the AI homework question at Caltech has no settled answer, and the university is letting each course find its own.

