Bitter Lesson misread: scaling won't brute-force biology, researcher argues
anshulkundaje · x · 2026-10-26
Rlacombe flags a widespread category error: invoking Sutton's Bitter Lesson to claim brute-force simulations will 'solve' biology. Sutton's actual point is subtler—winning architectures tend to be ones that scale with compute and data, like deep learning over hand-crafted features, or link-structure search over human curation. Simulation is neither learning nor search: it scales with compute but is hard to parallelize, missing out on GPU gains, and doesn't scale with data (weather observation assimilation aside). His punchline: don't parrot buzzwords you just learned.
More from AGI Musings
- Colleague writes flawless AI proposals but can't recall his own second point on a call — rvp · 2026-10-26
- Bill Gates warns AI will replace nearly all work skills: 'Hire the robot, fire the human' — sachinmaya1980 · 2026-10-26
- What scares AI researchers most? One tweet nails the bio-threat worry — imjustnewatai · 2026-10-26
- Philosophers Clash Over Whether Phronesis Is Useless in the AI Era — fkasummer · 2026-10-26
- Chip designer: AI hardware design is still just above the 'funny robot monkey tricks' stage — AccBalanced · 2026-10-26
- Harnesses for output variance may matter as much as model training — Afinetheorem · 2026-10-26