A finished homework sheet tells an instructor that something was submitted. Establishing whose understanding produced the answer is a harder question. UC Berkeley mathematician Richard Borcherds says growing access to AI has changed how he approaches that distinction.
In Zain Khan’s October 26 report based on a direct interview for The California Post, Borcherds says he now bases grades mainly on in-person exams without electronic devices. Homework still earns credit for completion, while correctness is no longer what he grades.
What the change actually covers
The account describes a change in assessment, rather than the disappearance of homework. Students can still receive credit for completing assignments. Borcherds argues that rewarding correct answers can encourage AI use and disadvantage students who work honestly on their own.
That is his explanation of the incentive problem. It is not a measured finding that every student cheats or that every submitted answer comes from a chatbot.
A higher average is an observation, not an individual diagnosis
Borcherds also describes a noticeable increase in average scores on online exams, while saying he cannot tell which students use AI. Both parts matter. A shift in scores cannot identify how each student produced an answer.
His response is to place more weight on device-free exams taken in person. The interview does not provide a controlled evaluation of whether that choice improves learning. It establishes the approach he says he is using and the problem he wants to address.
Keep the institutional boundary clear
This is one instructor’s account. It does not establish that UC Berkeley has abolished homework or that all its instructors now follow the same grading method. The affected course name, homework’s exact share of the final grade and the effect of the change are not verified in this package.
Ivy’s assessment
The useful question is what a task is meant to reveal. Practice can help students work through ideas; an assessment also needs a defensible basis for judging individual understanding. Those purposes deserve explicit attention when a correct answer is easy to obtain elsewhere.
An instructor could ask for a brief explanation of a solution or a related problem completed under supervision. These are editorial suggestions for making reasoning visible, not methods attributed to Borcherds or tested in this article. The concrete development here is narrower: he has changed which evidence he uses when awarding grades.
