When AI Helps Too Much: What Berkeley's Failing Grades Tell Us About Learning
Something concerning is happening in the computer science classrooms at UC Berkeley. In Spring 2026, failing grades in core CS courses like CS 10 and CS 61A jumped significantly, with CS 10 seeing a staggering 35.3% F rate and CS 61A at 10.6%. For context, these numbers rarely exceeded 10% in previous years, like Spring 2025 or Spring 2024. Even an upper-division course, EECS 127, hit a 16.8% F rate, well above the typical 5% guideline for such classes. Average GPAs in CS 10 and CS 61A dropped to C-pluses, around a 2.3 GPA. This isn't just a blip; it points to a deeper problem, and many educators are looking at how students are using large language models (LLMs) like Claude, ChatGPT, and Google Gemini, leading to a decline in AI usage math skills.
The Temptation of the Instant Answer
It's easy to see the appeal. You're stuck on a coding problem or a math proof, and an LLM can often spit out a plausible answer in seconds. For students facing tight deadlines or complex assignments, that's a powerful draw. Dan Garcia, who teaches CS 10 and CS 61A, believes this "vast increase in academic dishonesty" due to LLM usage is the primary driver behind the high failing rates. He's seen nearly 30 students in CS 10 caught cheating on take-home exams this semester alone. Students are leaning too hard on these tools, and it shows when they face exams without AI assistance.
This isn't just about cheating, though. It's about what happens to your brain when you outsource the hard work of problem-solving. On platforms like Hacker News, you'll find developers talking about a "cognitive decline" – a diminished ability to tackle problems without AI. The temptation to use AI for "speeding up" assignments often leads to a kind of "laziness" where foundational skills aren't truly internalized. While AI offers a performance boost, it often comes with "inevitable slop" that still needs manual correction, and if you don't understand the underlying concepts, you can't even spot the errors. This reliance on AI for immediate solutions, rather than as a learning aid, directly contributes to the observed decline in critical thinking and problem-solving abilities. Students are effectively bypassing the crucial mental exercises that build robust understanding, leading to a significant gap in their overall academic preparedness.
Why AI Usage Dwindles Math Skills
The problem extends beyond just coding. Gireeja Ranade, teaching EECS 127 (a course heavy on linear algebra and mathematical proofs), observed a "similar lack of prerequisite mathematical skills." Students are struggling with the core math concepts needed to succeed. When an LLM generates a proof or solves a complex equation, it gives you the answer, but it doesn't teach you the steps, the reasoning, or the intuition behind it. It's like using a calculator for every single arithmetic problem without ever learning how to add or multiply by hand. You get the right answer, but you don't build the mental model, which is crucial for developing strong AI usage math skills.
This isn't to say AI is inherently bad for learning. The challenge is how we use it. If you're just copying and pasting, you're bypassing the learning process. You're not "putting in the sweat" of learning, as professors like Ranade emphasize.
How to Use AI to Learn, Not Just Get Answers
So, what can we do? We can't put the genie back in the bottle. AI is here to stay, and it's a powerful tool. The key is to use it in a way that enhances learning, rather than replacing it.
- Use AI as a Tutor, Not an Answer Key: Instead of asking for the solution, ask the LLM to explain a concept, break down a complex problem into smaller steps, or generate practice problems. You could even ask it to critique your own solution, pointing out flaws in your logic or areas for improvement.
- Deliberate Practice with AI: After you've tried to solve a problem yourself, use the AI to check your work or compare approaches. If the AI gives you a solution, try to re-solve the problem without the AI, explaining each step to yourself. This builds muscle memory for critical thinking.
- Focus on "Why," Not Just "How": When an AI gives you code or a proof, don't just accept it. Ask "why" it chose that particular algorithm, "why" that mathematical step is valid, or "what are the trade-offs" of this approach. This pushes you beyond surface-level understanding.
- Educators Need to Adapt: As Ranade puts it, we need to teach students "more, not less" in the age of AI. This means designing assignments that require critical thinking, synthesis, and application of knowledge in ways that AI can't easily replicate. It also means being transparent about AI usage policies and educating students on ethical and effective AI integration. Some professors, like Garcia, plan to "advertise" the Spring 2026 issues to future classes and identify students needing remedial support. Over 1,300 UC faculty, including Garcia and Ranade, have even signed a petition calling for the reinstatement of ACT and SAT scores for STEM admissions, citing concerns about mathematical preparation. This proactive approach is essential to counter the negative trends observed and ensure students develop genuine proficiency.
The Path Forward
The rising failing grades at Berkeley are a stark reminder that technology, while powerful, is a tool. It amplifies our capabilities, but it doesn't replace the fundamental process of learning. For students, this means consciously engaging with the material, even when an LLM offers a shortcut. For educators, it means evolving our teaching methods to guide students in using these tools responsibly and effectively, ensuring they build the core skills that will serve them long after the latest AI trend fades.
The goal isn't to ban AI, but to integrate it thoughtfully. We need to teach students how to think critically with AI, not just let AI think for them. That's how we turn a potential academic crisis into an opportunity for deeper, more effective learning. The future of education at institutions like UC Berkeley hinges on this careful balance, ensuring that technological advancement supports, rather than undermines, the development of essential AI usage math skills and critical thinking.
