Interview
summary
Argues that real progress splits into learning world models separate from policy and tackling long-horizon reinforcement learning where sparse rewards force reliance on a good environment model; claims many products don’t need fancy deep research but linear regression on cleaned data and solid pipelines; frames the practical bottleneck as infrastructure and data engineering not algorithms; cites textbooks and industry examples rather than novel benchmarks or numbers.
AI-written summary · about
related talks
26mCompleting the machine learning loop
46mPlaying the long game
16mReal-World Robotics as the Next Frontier for AI?
06mInterview
27mPlatforms for valuable AI Products- Iteration, iteration, iteration
36mBand-Aids don’t fix bullet holes
33mAI with a DevOps mindset - experimentation, sharing ..
27mFireside Chat