The Forklift at the Gym: Why AI is Tanking Berkeley CS Grades

Imagine going to the gym, but instead of lifting the weights yourself, you use a forklift. The weights get moved successfully, but your muscles remain exactly the same. According to recent observations from one of the world's most prestigious universities, a similar phenomenon is currently playing out in computer science education, with artificial intelligence playing the role of the forklift.

At the University of California, Berkeley, professors in the Computer Science department are reporting a troubling trend: a significant spike in failing grades, accompanied by a noticeable deterioration in students' foundational math skills. The common denominator behind this academic slide appears to be the pervasive, unchecked use of artificial intelligence tools for coursework.

The core issue lies in the illusion of competence. AI coding assistants and large language models are incredibly efficient at debugging code, structuring algorithms, and solving complex mathematical equations. When students rely heavily on these tools to complete their daily assignments, they often bypass what educators call "productive struggle"—the frustrating, time-consuming, but neurologically essential process of wrestling with a difficult concept until it finally clicks.

Because AI delivers instant, polished answers, students reading these outputs often mistake the machine's competence for their own understanding. The reality check comes during closed-book, in-person exams. Stripped of their digital crutches, students who submitted flawless homework are suddenly finding themselves unable to work through basic foundational logic, leading to the soaring failure rates professors are now witnessing.

The decline in math skills is particularly alarming for computer science majors. In this field, mathematics is not merely about crunching numbers; it is the underlying language of algorithms, data structures, and systemic logic. If the next generation of software engineers outsources this foundational thinking to AI, they risk becoming mere assemblers of AI-generated code snippets rather than true architects capable of building or troubleshooting novel technologies. When a complex, undocumented system breaks in the real world, an AI trained on past data might not be able to fix it—and neither will an engineer who never learned the underlying math.

The takeaway from Berkeley is not that AI should be banned from the classroom. Such a move would be both impossible and counterproductive, given that these tools are now industry standards. Instead, this trend serves as a critical wake-up call for how we structure learning. The challenge for educators and students alike is to redefine the relationship with AI: transforming it from a surrogate that does the heavy lifting into a sparring partner that guides the workout. Only then can we ensure that as our tools get smarter, the humans wielding them do too.

Key Points

• UC Berkeley CS professors are seeing a spike in failing grades and declining math skills.
• Heavy reliance on AI tools for homework creates an "illusion of competence" that shatters during closed-book exams.
• Skipping the "productive struggle" prevents students from building essential neurological pathways for logic and problem-solving.
• The goal is not to ban AI, but to restructure education so students use it as a tutor rather than a surrogate.

Why It Matters

Understanding this trend is crucial for ensuring that the next generation of engineers actually learns the foundational logic required to build and fix complex systems, rather than just assembling AI-generated code.

Sources:

• Failing grades soar with AI usage, dwindling math skills in Berkeley CS classes — Hacker News AI top posts