Stanford CS312 teaches deep learning by experiment prediction; suggested as synthetic-data seed for frontier labs
jasondeanlee · x · 2026-11-22
Jason Dean Lee notes frontier labs are interested in hill-climbing on the science of deep learning and suggests seeding a synthetic data generator with quizzes from Stanford's new CS 312 "Deep Learning Alchemy" course. Taught Fall 2026 by Tatsunori Hashimoto and Suhas Kotha (CAs Kaiyue Wen, Tristan Thrush), with materials public, the 5-unit course argues math knowledge alone can't produce the next generation of architecture inventors: students run many experiments in tractable domains, complete experiment assignments, and take seven difficult quizzes testing their ability to predict held-out experiment outcomes, with no inflated grading curve.
More from Companies & People
- Gary Marcus predicts OpenAI won't IPO and will instead sell at a discount to a Mag7 company — GaryMarcus · 2026-11-22
- Gary Marcus warns Anthropic bulls: Lean-verified math wins won't generalize — GaryMarcus · 2026-11-22
- Neon backed by Germany's largest seed round ever, says Scobleizer — Scobleizer · 2026-11-22
- Team says Grok Bot agents ran half the prep for a 5,000-person summit in months — GavinSBaker · 2026-11-22
- Dubai courts global IT firms with City AI Agent push for public services — nordicinst · 2026-11-22
- Max Welling interview: AI, CuspAI and parallels between AI and quantum physics — NandoDF · 2026-11-22