Stanford's new CS312 'Deep Learning Alchemy' course teaches experiment design; one user suggests feeding its quizzes to synthetic data pipelines

jasondeanlee · x · 2026-11-22

Stanford is launching CS312: Deep Learning Alchemy (Fall, taught by Tatsunori Hashimoto and Suhas Kotha), with all recordings and materials to be made public. It's a non-standard deep learning intro that skips fundamentals like backprop and instead trains experimental skill: students run many computationally tractable experiments and take seven hard quizzes testing their ability to predict held-out experiment outcomes. It's a 5-unit class with no grade inflation.

The poster suggests labs feed the course's quizzes and homework into synthetic data generators and hill-climb on them, arguing this could build "good science of DL" abilities for pretraining—a mini-RSI recipe he suspects labs already run.

Related event: Stanford's Deep Learning Alchemy course sparks synthetic data ideas(3 posts)→

Original post →

More from Companies & People

Companies & People channel →