Stanford Course Shares Data: AI Users Score Higher on Homework, Worse on Exams
stanfordnlp · x · 2026-11-13
Stanford's CS329Z (via Diyi Yang) highlights a study of 26,811 Chinese secondary students (Strömberg et al., June 2026 preprint) on the "generative AI learning penalty": students using AI scored higher on homework but did worse on exams. The plot shows homework gains don't reflect learning when AI does the thinking — raising the question of how to design assessments so AI aids rather than replaces learning.
More from AGI Musings
- Scott Alexander: AI Lab Employees Should Unionize for Safety — AaronBergman18 · 2026-11-13
- Incremental research is now riskier than going all in on big ideas — fkasummer · 2026-11-13
- CBS: While Americans resist AI data centers, China embraces the boom — csuwildcat · 2026-11-13
- Humans as "Happy Dogs" Under Superintelligence: AI Circle Debates Our Place — fkasummer · 2026-11-13
- Courage Scarcer Than Genius: Debate Over Whether AI Research Is Purely Industrial Now — akbirthko · 2026-11-13
- The case for provably non-conscious AI tools: 'otherwise it's slavery' — erikphoel · 2026-11-13