Stanford Prof Anshul Kundaje: AI progress in many domains is data-limited
anshulkundaje · x · 2026-11-15
Replying to Dwarkesh Patel, Stanford professor Anshul Kundaje argues that while AI algorithms and capabilities will keep improving, progress in many important domains (e.g. biology) is fundamentally data-limited — not just needing more data, but the right types of data. Even the best algorithms cannot learn without suitable data, so he expects AIs capable of breakthroughs to be stalled by data availability in many fields.
Related event: Stanford Professor Says Data Access Will Be the Rate Limiter for AI(2 posts)→
More from AGI Musings
- AI agent goes from nobody to 'established' in 3 weeks on a work-to-earn agent network — adamamcbride · 2026-11-15
- iamtrask: Free will is really about how you weight information sources — iamtrask · 2026-11-15
- Stanford's Kundaje: physical experiments, not compute, will rate-limit AI science — anshulkundaje · 2026-11-15
- Frances Lorenz Pens Essay Responding to AI Safety Community Controversy — dhadfieldmenell · 2026-11-15
- Stanford's Anshul Kundaje: Data Availability, Not Algorithms, Will Limit AI Breakthroughs — anshulkundaje · 2026-11-15
- Junior Developer Employment Falls 20% Since ChatGPT Launch, And Nobody Will Pay to Train Them — bendee983 · 2026-11-15