UC Berkeley Computer Science Failing Rates Spike Amid AI Usage and Math Skill Decline

Academic Performance Decline in UC Berkeley CS Courses

Failing grades in several UC Berkeley computer science courses surged in Spring 2026, significantly exceeding departmental grading guidelines. According to data from Berkeleytime, 35.3% of students in CS 10 ("The Beauty and Joy of Computing") and 10.6% of students in CS 61A ("The Structure and Interpretation of Computer Programs") received failing grades.

This marks a sharp increase from Spring 2024 and 2025, where failure rates for both courses remained below 10%. These figures deviate heavily from the Electrical Engineering and Computer Sciences (EECS) department's guidelines, which suggest that only 7% of students in lower-division courses should receive D or F grades. Additionally, the average GPA for these classes dropped to 2.3 (C-plus), well below the typical range of 2.8 to 3.3.

The Role of Generative AI and Academic Dishonesty

Teaching professor Dan Garcia identifies a "vast increase in academic dishonesty" driven by Large Language Models (LLMs) like Claude, ChatGPT, and Google Gemini as the primary driver of these failure rates. The impact manifests in two distinct ways:

1. Direct Prosecution: A significant portion of failing grades resulted from students being caught cheating and referred to the center for student conduct. In CS 10 alone, nearly 30 students were caught cheating on take-home exams.
2. Skill Atrophy: Many students rely too heavily on LLMs to complete assignments, leaving them unprepared for high-stakes exams where AI assistance is unavailable.

Professor Garcia argues against the use of grade curves, which he believes hide these systemic issues. He advocates for clear, public point thresholds for letter grades, allowing students multiple opportunities to reach those standards without lowering the bar.

Dwindling Mathematical Preparedness

Beyond AI usage, faculty report a critical lack of prerequisite mathematical skills among incoming students. Associate teaching professor Gireeja Ranade observed a 16.8% failure rate in EECS 127 ("Optimization Models in Engineering"), compared to the typical 5% for upper-division courses.

Ranade noted that students struggling with linear algebra in office hours revealed that some prerequisite courses at UC Berkeley had adopted "open-internet, open-AI policies" for homework and exams, potentially undermining the mastery of foundational concepts.

In response to these trends, Garcia and Ranade joined over 1,300 UC faculty in signing a petition to reinstate ACT and SAT standardized testing scores for STEM admissions, arguing that test-free admissions have failed to reliably assess student readiness.

Institutional Challenges and Student Engagement

Staffing shortages and declining engagement have further complicated the educational environment:

• Reduced Support: Due to high hourly wages for TAs, the EECS department has reduced both undergraduate enrollment and the number of undergraduate TAs. This forced Professor Ranade to remove a final project from EECS 127, a component that typically yielded high scores for students.
• Engagement Drop: Both Garcia and Ranade reported a surprising decline in office hour attendance. Garcia noted that after years of full office hours, he found himself sitting alone, signaling a disconnect between students and instructional support.

Synthesis of Community Perspectives

Discussion among technical professionals and alumni suggests that the Berkeley experience is part of a broader cognitive shift.

The "Cognitive Decline" Hypothesis

Several observers argue that LLMs are creating a dependency that erodes deep thinking and brainstorming capabilities. One commenter noted that even PhD-level professionals are struggling to code or write without an LLM, suggesting that the "cognitive decline" is extending beyond the classroom into professional environments.

The Debate on Admissions and Pedagogy

While some attribute the failure rates to AI, others argue that the removal of standardized testing is the root cause, claiming that universities are now admitting students who lack middle-school level math proficiency. Others suggest that the problem lies in outdated pedagogy, arguing that traditional lecture-and-test models are obsolete in the AI era.

AI as a Learning Tool vs. Shortcut

There is a consensus that AI can be a powerful pedagogical tool if used for Socratic dialogue or focused drilling, but it becomes detrimental when used to bypass the "struggle" of learning. As Professor Garcia summarized, "Confusion is the sweat of learning," and many students are now avoiding that effort.