Failing grades soar with AI usage, dwindling math skills in Berkeley CS classes
UC Berkeley CS failing grades tripled in spring 2026, with professors citing AI overuse, weak math skills, and understaffing.
Failing grades in UC Berkeley computer science courses spiked dramatically in spring 2026. CS 10 saw 35.3% F's and CS 61A saw 10.6% F's, compared to under 10% in prior years, well above the department's 7% guideline. Teaching professor Dan Garcia caught nearly 30 CS 10 students cheating via LLMs on take-home exams and estimates many more relied heavily on AI tools like ChatGPT and Claude without learning underlying concepts. Professor Gireeja Ranade's upper-division EECS 127 course saw 16.8% F's, partly due to students lacking prerequisites in linear algebra and vector calculus. Both professors also cited understaffing reducing support and office hour attendance. Garcia opposes grade curving and advocates for transparent thresholds; Ranade emphasizes teaching deeper critical thinking in the AI era.
What HN community is saying
The thread's dominant concern is title ambiguity: "failing grades soar" could mean failing grades increased (the actual reading) or failing students improved, though commenters debate whether this ambiguity is accidental or clickbait. A substantive fork emerges over AI's role in education. One camp argues LLMs fundamentally misalign with learning goals, producing plausible-sounding text without understanding. The opposing view contends AI can complement self-directed learning when used alongside textbooks and practice, citing personal experience with learning graphics programming and noting that some professors hallucinate worse than current LLMs. Both camps agree institutional enforcement and student motivation are critical variables independent of AI.