Tatsu Hashimoto on teaching DL: teach the standard mental model, then let students break it
srush_nlp · x · 2026-10-29
Tatsu Hashimoto shares his deep learning pedagogy: empirical DL is messy enough that students must experience failing to predict model behavior firsthand, so prediction quizzes set the right incentives. Yet theories with local predictive value are worth lecturing on. His approach: spend most lectures on the 'standard mental model,' then push students to mess with it in assignments. Knowledge still matters — it's just hard to internalize properly.
More from Companies & People
- Microsoft execs detail how accelerators, codegen and custom agents transform internal dev — kevin_scott · 2026-10-29
- Atlassian Signs On as Premium Partner for AI Engineer Sydney, Dec 2026 — davidhoang · 2026-10-29
- Elon Musk calls Google's Gemini 4 Carbon "impressive" after benchmark-topping launch — vedantmisra · 2026-10-29
- AI streamer Mika tries to launch a pump coin, tweets the dev wallet's private key instead — econoar · 2026-10-29
- Reddit mocks SSI: 'bubble about to burst' and Ilya still hasn't shipped a model — Snoo26837 · 2026-10-29
- Non-technical roles at top AI labs pay up to $460K: OpenAI, Anthropic, Thinking Machines openings listed — lennysan · 2026-10-29