Stanford launches CS312 Deep Learning Alchemy, a course built on experiment quizzes for LLM pretraining
CatAstro_Piyush · x · 2026-11-13
Stanford's new Fall 2026 course CS312: Deep Learning Alchemy, taught by Tatsunori Hashimoto and Suhas Kotha, targets aspiring LLM pretraining researchers. Its core thesis: math alone isn't enough — intuition comes from running many experiments and learning to predict held-out outcomes.
- Format alternates assignments (running experiments), quizzes (predicting experiment results), lectures and tutorials; all recordings and materials will be public.
- 5 units, no inflated grading curve, lots of experiments plus seven hard quizzes.
- Sample quiz from the course: rank six Transformer ablations (no prenorm/residuals/MLP/attention/RoPE) by final validation loss across peak learning rates.
More from Companies & People
- Luminai Topology Launches, Cutting Cleveland Clinic Referral Processing From 3+ Days to Same-Day — ycombinator · 2026-11-13
- BOLD Lab opens applications for PhD students starting in 2027 — j_foerst · 2026-11-13
- Tech Leaders Collectively Defend Scandal-Plagued CEO, Mocked for 'Wanting Crucifixion' — zacharynado · 2026-11-13
- AI study chatbots: $72 vs under $5 per student across business schools — joshgans · 2026-11-13
- Sentry CEO David Cramer mocks agile estimation: 'imagine still using story points' — zeeg · 2026-11-13
- Packed room at AIE Code Summit for Zach Lloyd on evals and self-improving agent loops — vikvang1 · 2026-11-13