Chinese pupils using AI raised homework scores 18% but scored 20% lower on exams
- A study tracking 27,000 Chinese pupils aged 12 to 18 found that students using AI scored 20% below non-users on exams after six months, despite stronger homework results.
- AI users' average homework scores rose 18% across subjects, while average completion time fell from 64 minutes to 45 minutes per assignment.
- About 80% of the tracked pupils reported using models including Doubao and DeepSeek; the remaining 20% formed the study's control group.
- The study was conducted by David Stromberg of Stockholm University with Victor Lei and Wu Yanhui of the University of Hong Kong.
- Chegg's survey found 80% of rich-world undergraduates used AI in their studies last year; more recent polls put use at 94% in Britain and 93% in Germany.
Hacker News opinions
Homework is supposed to build understanding, so the process matters more than the submitted answer. AI could explain concepts or adapt questions to a student's knowledge, but it needs guardrails that stop it from simply writing the answer.
If students use AI for homework, they finish faster and get higher grades, then get crushed on the exam.
I read the result differently: AI users who studied about as long as high-performing non-users did slightly better. The students who did badly were using it to do the homework for them, so AI looks like an amplifier of both good and bad study habits.
I am skeptical of that optimistic interpretation. The claimed slightly higher-performance group is statistically insignificant, only 4 to 20 students depending on context, while 81% of AI users were classified as outsourcing homework and that rate rose with more LLM exposure.
I want to know which students reach for AI and why. School funding, device access, textbooks, curriculum limits, and teacher incentives will shape what
headline
importance
keyFindings
slug