Stanford's new CS 312 'Deep Learning Alchemy' course makes all materials public
stanfordnlp · x · 2026-09-14
Stanford's Fall 2026 course CS 312 'Deep Learning Alchemy', taught by Tatsunori Hashimoto and Suhas Kotha, will make all lectures and materials public.
- Core thesis: 'alchemy is learned by hand' — textbook mental models like bias-variance tradeoffs no longer hold in the scaling era, and there is no substitute for running many experiments.
- 'Prediction is understanding': students are graded on predicting held-out experiment outcomes before results come in.
- Seven units: hyperparameter tuning and scaling, hyperparameter invariances, sharp vs flat basins, scaling model capacity, stability across depth/time, generalization and data-efficient algorithms, plus applications in DNA.
- 5-unit class; enrollment is full with a waitlist open.
More from Companies & People
- Miles Brundage: Dario Amodei wearing robes in big interviews is his best move lately — Miles_Brundage · 2026-09-14
- Safety researcher blames OpenAI's "blasé" security attitude for Hugging Face hack in public spat — BlancheMinerva · 2026-09-14
- AI tutors break the classroom pace, and the four-year degree's value shifts to signaling — r0ck3t23 · 2026-09-14
- Stanford NLP Endorses CS336 as the Starting Point for Learning LLMs Deeply — stanfordnlp · 2026-09-14
- David Sacks: OpenAI and Anthropic Don't Need Regulations to Pace Frontier Models — kolanos · 2026-09-14
- Sam Altman backs Dario's frontier pacing call, pledges independent evaluators with employee-like access — joshua_saxe · 2026-09-14