Stanford's new Deep Learning Alchemy class: 7 exams worth 85% of the grade
tatsu_hashimoto · x · 2026-10-14
Stanford CS312 'Deep Learning Alchemy', taught by Tatsu Hashimoto and Haslam Kothasu, teaches deep learning the hands-on way: students build intuition by running experiments on small LMs (e.g. residual scaling and noise at different learning rates), then take closed-book prediction exams — given an unseen code diff, predict the effect on validation loss. Grading is unconventional: 7 exams count for 85%, with 15% in-person discussion assessment. Students may use AI agents for experiments and assignments but must demonstrate mastery on exams.
Related event: Stanford's Deep Learning Alchemy course sparks debate(4 posts)→
More from Companies & People
- PartyKit shuts down free hosted platform 2.5 years after Cloudflare acquisition — threepointone · 2026-10-14
- OpenAI races to ship next flagship model amid fallout from safety researcher firings — steph_palazzolo · 2026-10-14
- Carmack's 'give a damn' quote fuels backlash against AI-slop creative software — teropa · 2026-10-14
- NVIDIA Research Hiring Scientists and Interns for World Models, Diffusion LMs, Protein Design — chaitjo · 2026-10-14
- Anthropic launches its first Japan-wide online Claude Code hackathon with $50K in credits — ClaudeDevs · 2026-10-14
- Spawn runs 8 people on $7k/day of agents, out-pacing Anthropic per capita — mattshumer_ · 2026-10-14