Stanford's new CS 312 makes students build DL theory with Claude via experiment prediction
srush_nlp · x · 2026-10-29
Stanford unveiled CS 312: Deep Learning Alchemy (Fall 2026), taught by Tatsunori Hashimoto, Suhas Kotha and others, with all materials and recordings to be made public.
- The course's stance: knowledge or math alone isn't enough — students must run many experiments, with the burden of theory construction placed on the students (and their Claude).
- Format: alternating assignments (running experiments to understand deep learning phenomena) and quizzes (predicting held-out experiment outcomes), seven hard quizzes, no inflated grading curve.
- srush calls the design "transformative and surprisingly challenging — maybe knowledge in ML was never that critical."
More from Companies & People
- GitHub launches Copilot Fusion as CPO frames AI strategy: lower the floor, raise the ceiling — mariorod1 · 2026-10-29
- miHoYo publicly questions DeepSeek's approach, betting AGI on hard problems not token economics — teortaxesTex · 2026-10-29
- As Microsoft offers 60,000 employees the door, a founder shares pre-seed startup advice — MartinGTobias · 2026-10-29
- First touchscreen Mac revealed, andrewchen declares monitor-touching 'an abomination' — andrewchen · 2026-10-29
- Every's weekly: one headline rewrite took agent citations from 2/240 to 240/240 — every · 2026-10-29
- Mercor hits $2B ARR just 4 months after $1B, and its engineers now build with Devin — silasalberti · 2026-10-29