The article reports that failing grades spiked in a handful of UC Berkeley computer science courses, especially introductory ones, and quotes instructors who see two linked problems. Students are using AI to complete homework and take-home exams, and many are arriving with weaker mathematical foundations than those courses assume. One cited example was a prerequisite linear algebra class that reportedly allowed open internet and open AI use on both homework and exams. Berkeley instructors also pushed back on grade curves, arguing that curving hides a real drop in mastery and turns a standards problem into a sorting exercise.
Most comments landed on a blunt point: the issue is not that AI exists, it is that students are using it to skip the hard reps that make knowledge stick. People drew the same distinction over and over. LLMs can work as tutors, sparring partners, quiz generators, worksheet builders, or critics of your own draft. They are destructive when used as answer machines. Several commenters said they feel this in their own work already. They can ship more with coding agents, but their patience, attention span, memory for details, and willingness to wrestle with difficult problems all degrade if they let the model carry too much of the load. The loss people worry about is less factual recall than the ability to sit with confusion long enough to build new mental models.
The strongest pushback was against the article’s framing, not the broader concern. Several commenters noted that the piece appears to cherry-pick a few Berkeley classes and small cohorts, especially spring sections of intro courses, rather than showing a campus-wide rise across all
CS offerings. Others argued the article mixes together three different failures: students being caught cheating, students arriving underprepared in math, and students failing in-person assessments after outsourcing practice to AI. A large side debate focused on UC admissions policy. Many commenters think dropping
SAT and
ACT requirements weakened the math floor for
STEM entrants, while others said that cannot explain a sudden 2026 jump and that the immediate change is better explained by AI becoming good enough in the last two years to let students coast through prerequisites. COVID learning loss, long-running declines in attention, and anti-intellectual drift were also treated as contributing factors.
Where the comments ended up is pretty practical. Homework can no longer be trusted as a measurement tool. Ungraded homework, frequent quizzes, oral exams, in-person practical tests, and assignments where students must explain tradeoffs are all more credible than take-home deliverables. For working engineers, the same rule applies. Using LLMs after you already know the domain can be a real force multiplier. Using them before you have the basics turns you into a code reviewer of work you barely understand. The thread was gloomy about current incentives, but not confused about the fix: if the goal is learning, the system has to reward doing the thinking, not merely submitting something that passes.