Stanford's new CS 312 grades 85% on predicting experiment outcomes, 0% on homework
stanfordnlp · x · 2026-09-13
Stanford NLP launched CS 312: Deep Learning Alchemy by Tatsunori Hashimoto, built on the premise that prediction equals understanding: assignments count for 0% of the grade, while 85% comes from 40-minute quizzes where students see a code diff for an experiment they've never run and must predict validation loss differences. All materials are public, in the style of CS336.
More from Companies & People
- Sam Altman publicly agrees with Dario on pacing the frontier, commits to independent evaluators with employee-like access — rand_longevity · 2026-09-13
- Nadella: the next AI moat isn't the model — it's the learning loop only your company can run — rohanpaul_ai · 2026-09-13
- Sentdex asks: is there a single non-EA AI evaluation company frontier labs actually use? — Sentdex · 2026-09-13
- Rob LeCerc on lab cartel claims: all frontier labs seeded by EA, even Elon backs a slowdown — room left for an e/acc lab? — robleclerc · 2026-09-13
- AI pacing was supposed to slow capital, yet firms race to IPO anyway — xiaosun86 · 2026-09-13
- September AI recap: machine-proved Fermat, GPT 6 Astra, state probes and a superintelligence ban push — peterwildeford · 2026-09-13