Stanford Prof Bets LLMs Reach Navier-Stokes Level in 2 Years; Biology Researcher Pushes Back
anshulkundaje · x · 2026-09-09
Stefanos Batzoglou recounts how his planned "LLMs can't reason" paper—thousands of trivially easy reasoning tasks that top models scored 10% on—was fully saturated by end of 2025, with the field now eyeing Navier-Stokes; he bets the landscape will shift just as much in two years.
Computational biologist Anshul Kundaje pushes back: models remain limited by the nature of trainable data, and on hard biology problems lacking necessary data, results are sobering. He argues for discussing what models can do today based on evidence rather than hunches, especially for opaque frontier models.
More from AGI Musings
- Ex-OpenAI VP Miles Brundage: if you're considering leaving the AI industry, you probably should — Miles_Brundage · 2026-09-09
- Senate Tech Committee Schedules Zero AI Hearings for Next 4 Months — austinc3301 · 2026-09-09
- 'Attach sensors, train RL on the real world — then we're done': a one-line AGI thesis — Paimaamu · 2026-09-09
- AI 2027 was mocked as too fast — GPT-6 Astra just landed on its curve — Confident_Salt_8108 · 2026-09-09
- Compute capacity is now the great divider, says engineer Burcu Dogan — rakyll · 2026-09-09
- Anthropic researcher: >10% chance AI kills all humans within a decade, no alignment plan yet — Delahuntagram · 2026-09-09