Karpathy says AI‑homework crackdown failed, urges in‑class grading shift
Andrej Karpathy - the guy who used to run Tesla’s AI team - isn’t shy about the new school rules trying to flag AI-written work. He says the “war on AI homework” isn’t really winning; it’s pricey and, frankly, not that effective. Sure, schools want to protect academic honesty, but students seem to find ways around the detectors fast, leaving teachers with a shaky sense of safety.
Karpathy thinks we should stop leaning on after-the-fact checks and instead do assessments while a teacher watches. That would make it tougher for learners to lean on language models during a test and might push them back toward solving problems on their own. The takeaway?
If schools care about genuine learning, they’ll have to rethink grading.
So, Karpathy argues most grading should move into the classroom, where teachers can actually see what students are doing. That way, kids stay motivated to figure things out without AI, knowing they’ll be evaluated in a setting where shortcuts aren’t an option. He stresses that he’s n
As a result, Karpathy believes the majority of grading has to shift to in-class work, where teachers can physically monitor students. This change keeps students motivated to learn how to solve problems without AI, since they know they will be evaluated without it later. Karpathy stresses that he's not pushing for an anti-tech school system. Students should learn to use AI because it is "here to stay and it is extremely powerful." As he puts it, no one wants students to be "naked in the world" without access to it.
Karpathy says schools are already behind a fast-moving AI tool. He thinks the detection arms race is lost, so the only sensible move is to shift most assessments into the classroom where teachers can actually see students work. In that setting, learners might stay motivated to solve problems instead of slipping hidden help.
Still, how to redesign curricula, train staff and carve out class time is still fuzzy. Some critics worry constant monitoring will stretch resources and change how teachers teach. The idea that physical supervision is the only reliable guard hasn’t been tried on a large scale, so it’s hard to know if it will hold up.
Whether schools can adopt this model without hurting other goals remains an open question. For now the talk has moved from trying to catch AI use to rethinking assessment altogether, leaving educators to judge if in-class grading is practical against the ongoing lure of AI-generated work.
Related Reading
Research & Benchmarks Consensus uses GPT-5 and Responses API to speed scientific research
Research & Benchmarks Developers say Sora, unlike Vine/TikTok, is not about people in social media
Research & Benchmarks Google AI Advisors Let Users Probe Performance with Conversational “Why” Queries
Common Questions Answered
Why does Andrej Karpathy claim the AI‑homework crackdown has failed?
Karpathy argues that the crackdown is costly and ineffective because students quickly learn to bypass detection tools, giving teachers a false sense of security. The policies have not prevented AI‑generated assignments, undermining the intended preservation of academic integrity.
What solution does Karpathy propose to address the shortcomings of AI‑generated assignments?
He suggests shifting the majority of grading to in‑class work where teachers can physically monitor students. This approach keeps learners motivated to solve problems without AI assistance, as assessments would occur under direct observation.
How does Karpathy reconcile the use of AI with his recommendation for in‑class grading?
Karpathy emphasizes that he is not advocating an anti‑technology stance; instead, he wants students to learn to use AI responsibly because it is powerful and here to stay. In‑class grading ensures students develop problem‑solving skills while still being exposed to AI tools outside the classroom.
What challenges does Karpathy acknowledge in moving assessments into the classroom?
He notes that redesigning curricula, training staff, and allocating classroom space present logistical hurdles. Despite these challenges, he believes the shift is necessary to stay ahead of the rapidly advancing AI tools.