Published: June 04, 2026, 15:29 UTC
The numbers are hard to ignore. In spring 2026, multiple computer science courses at the University of California, Berkeley recorded failing grades at rates significantly higher than previous semesters. Average grades dropped to C-plus territory, a GPA of roughly 2.3, well below the department’s own guideline of 2.8 to 3.3.
The Daily Californian, Berkeley’s student newspaper, reported the trend and what professors believe is driving it: increased AI usage among students, combined with eroding foundational math skills. The story quickly went viral on Hacker News, where it accumulated 489 points and more than 400 comments within hours.
The data
The grade shifts are concentrated in lower-division CS courses, the gateway classes that determine whether students proceed in the major. Berkeley’s grading guidelines explicitly state that “a typical GPA for a lower division course will fall in the range 2.8 to 3.3.” Spring 2026 saw both surveyed classes land at 2.3, a C-plus average.
Professors cited two converging factors. The first is widespread use of AI tools to complete assignments. Students submit code and problem sets generated by large language models without understanding the underlying concepts, then fail exams that test the material directly. The second is a decline in mathematical preparation; students arrive with weaker algebra and calculus foundations than previous cohorts, making it harder to grasp the theoretical material that CS coursework builds on.
The wider debate
The Berkeley story has become a flashpoint in a broader argument about AI in education. Sridhar Vembu, CEO of Zoho, weighed in on X with a warning: “AI can make you smarter faster, but AI can also make you dumber faster.”
The Hacker News discussion captures the tension well. One commenter with sympathy for the students noted that if large language models had been available during their own education, they would have used them too, then failed their exams. Another observed a subtler phenomenon among colleagues: even PhD-level researchers appear to be losing the ability to think through problems without an LLM doing most of the work. “Many of them can no longer sit quietly for even 30 minutes just thinking on their own,” the comment read, “which is a required skill for producing original thought.”
That comment struck a nerve. It received dozens of replies, some pushing back on the claim that cognitive decline is measurable, others agreeing that the pattern is visible and accelerating.
Not just Berkeley
Berkeley is not an isolated case. The pattern has been observed at other universities, although Berkeley’s status as one of the world’s top CS programs gives the story particular weight. If students at one of the most selective public universities in the United States are struggling this way, the implication is that the problem is structural, not institutional.
The question raised by the Berkeley data has no clean answer. Banning AI tools in the classroom is impractical when those same tools are essential in industry. Allowing unrestricted use risks graduating students who can operate an LLM but cannot design an algorithm.
Some faculty members are calling for changes to admissions criteria, placing greater emphasis on mathematical readiness. Others argue that curricula need to be redesigned around the reality of AI-assisted work, testing for understanding rather than output. Neither approach is straightforward, and both carry trade-offs that departments are only beginning to grapple with.
What comes next
Berkeley’s CS department is one of the most influential in the world. Its response to this grade crisis will be watched closely by every other institution facing the same dilemma. The data from spring 2026 suggests that the current approach is not working. The question is what replaces it.
Sources: The Daily Californian (June 3, 2026); Hacker News discussion (June 4, 2026); India Today (June 4, 2026)