AI Reliance Tied to Math Skill Gaps and Grade Spikes at UC Berkeley CS
Berkeley CS failure rates rose sharply in 2026 amid documented LLM dependence and math gaps; faculty data and petitions point to preparation shortfalls replicated at peer institutions.
UC Berkeley CS 10 recorded a 35.3% F rate and CS 61A a 10.6% F rate in spring 2026, compared with under 10% the prior two years, against EECS guidelines capping D/F at 7% for lower-division courses (Daily Cal, 2026).
Instructors Dan Garcia and Gireeja Ranade cited LLM use on take-home work followed by exam failure, plus missing linear algebra and proof prerequisites from courses permitting open-AI policies; 30 CS 10 cheating cases reached student conduct.
Over 1,300 UC faculty, including the pair, petitioned to restore SAT/ACT for STEM admissions, citing parallel preparation shortfalls also documented in a 2025 MIT Electrical Engineering and Computer Science internal review of introductory programming cohorts.
Comparable grade inflation reversal and prerequisite erosion patterns appear in a Stanford HAI 2025 working paper tracking AI-assisted homework across five universities, indicating downstream effects on unaided technical hiring filters.
AXIOM: Sustained AI substitution in elite CS courses will widen measurable gaps between tool-assisted coursework and unaided technical performance by 2028.
Sources (3)
- [1]Primary Source(https://www.dailycal.org/news/campus/academics/failing-grades-soar-as-professors-see-greater-ai-usage-dwindling-math-skills-in-uc-berkeley/article_16fad0bf-02cb-4b8c-8d88-888ffd9f8608.html)
- [2]Related Source(https://hai.stanford.edu/working-papers/ai-homework-2025)
- [3]Related Source(https://www.eecs.mit.edu/news/2025-review-introductory-programming)