Stanford's new CS312 makes assignments worth 0%, grades you on predicting loss curves
PMinervini · x · 2026-09-13
Stanford NLP has launched CS 312: Deep Learning Alchemy, designed by Tatsunori Hashimoto, with all materials public in the cs336 style.
The core premise is "prediction == understanding":
- Assignments count for 0% of the grade
- 85% of the grade comes from 40-minute quizzes where you're shown a code diff of an experiment you've never run and must predict validation loss differences
- The logic: without intuition for an experiment's outcome, your understanding is surface-level
The poster praising it notes how easy it would be to teach the same 2022 AI class today — instead, this professor built a course that tests real research intuition while keeping up with the frontier.
More from Companies & People
- Bittensor subnet SN44 October recap: major real-time graphics company testing its football model — richdotca · 2026-10-31
- Talk: how AI is reshaping education, from edtech builders to university classrooms — Cohere · 2026-10-31
- Google: AI Mode, AI Overviews and Search merging into one unified product — gaganghotra_ · 2026-10-31
- HubSpot co-founder: startups die from suicide, not homicide — competition isn't the real killer — nathanbenaich · 2026-10-31
- Anthropic culture anecdote: co-founder balked at office sushi, 'I'd prefer rice and beans' — himanshustwts · 2026-10-31
- Meta exec: Llama 3.2 was the first named model to run in space, back in April 2025 — Ahmad_Al_Dahle · 2026-10-31