UC Berkeley CS instructors link grade slump to AI reliance and math gaps
Data from Berkeleytime reveals failing grades in CS 10 and CS 61A reached 35.3 per cent and 10.6 per cent respectively, as educators cite academic dishonesty and inadequate mathematical preparation.
Computer science instructors at the University of California, Berkeley, are reporting a significant deterioration in student performance during the spring 2026 semester, attributing the trend to increased reliance on large language models and insufficient mathematical foundations. According to data published by the independent student-run newspaper Berkeleytime, failing grades in introductory and intermediate courses have surged well beyond historical norms and departmental expectations.
The Electrical Engineering and Computer Sciences department’s grading guidelines typically anticipate that 7 per cent of students in lower-division courses will receive D or F grades, with average GPUs expected to fall between 2.8 and 3.3. In spring 2026, however, CS 10, titled “The Beauty and Joy of Computing,” saw a 35.3 per cent failure rate, while CS 61A, “The Structure and Interpretation of Computer Programs,” recorded a 10.6 per cent failure rate. Both courses averaged C-plus grades, corresponding to a 2.3 GPA, a marked departure from the 10 per cent or lower failure rates observed in the preceding two years.
Teaching professor Dan Garcia identified a “vast increase in academic dishonesty” as a primary driver of these statistics, linking it directly to student usage of tools such as ChatGPT, Claude, and Google Gemini. Garcia reported that nearly 30 students in CS 10 were caught cheating on take-home exams, with cases referred to the centre for student conduct. He noted that while some failures resulted from prosecuted misconduct, others stemmed from students who relied too heavily on AI assistance and were unprepared for examinations conducted without such aids.
The issue extends to upper-division coursework, where associate teaching professor Gireeja Ranade observed similar struggles in EECS 127, “Optimization Models in Engineering.” The course recorded a 16.8 per cent F rate, significantly higher than the 5 per cent typical for upper-division classes. Ranade highlighted that students entering the course were expected to possess proficiency in linear algebra, vector calculus, and mathematical proofs, yet many lacked these prerequisite skills. She noted that some students had taken previous linear algebra courses with open-internet and open-AI policies, which failed to build the necessary analytical rigour.
In response to these preparation gaps, Garcia and Ranade have joined more than 1,300 UC faculty members in signing a petition calling for the reinstatement of SAT and ACT scores for STEM admissions within the university system. The petition argues that standardized testing is necessary to ensure incoming students possess the mathematical baseline required for rigorous computer science curricula. Department chair Jelani Nelson confirmed via X that the campus has reduced both undergraduate CS enrollment and teaching assistant numbers due to the high hourly wages associated with undergraduate staff.
Beyond academic performance, instructors reported a decline in student engagement, with office hours that were previously overflowing now seeing very low attendance. Garcia described sitting alone in his office for the first time, while Ranade noted a stark contrast to previous semesters. Both educators are reconsidering their pedagogical approaches, with Garcia planning to explicitly address the consequences of AI misuse in future classes and Ranade emphasizing the need to teach critical thinking skills that AI cannot replicate.
The grading structure in Garcia’s courses did not utilise curves, with letter grades determined by fixed point thresholds. Garcia argued against the practice of capping the number of A grades, a policy implemented at institutions such as Harvard, stating that clear standards should be maintained while providing students with ample opportunity to meet them. He expressed concern that curving masks underlying issues, suggesting that the current grade distribution reflects a genuine deficit in student readiness rather than a need for grade inflation.
Ranade also modified the structure of EECS 127 this semester, removing a final project component due to understaffing. She noted that this project typically yielded high scores, and its removal contributed to the lower overall grades. Despite the challenges, both professors underscored the importance of preparing students for long-term leadership roles, with Ranade stressing that educators must teach students to embrace confusion and difficult problems as part of the learning process.
