Writing · The Meta Gradient and elsewhere
Working it out in public.
Essays on world models, learning and data. Each one pairs the derivation and the code with the picture that made it click for me.
Subscribe on Substack →Research blogs
Interpretability & safety
Learning & world models
- The Architecture of Dreams, Part 3: Does Looking Right Mean Understanding?Genie, V-JEPA 2, and Physics-IQ.
- The Architecture of Dreams, Part 2: What Should a World Model Predict?MuZero, DreamerV3, TD-MPC2.
- Freezing the NoiseHow one line of algebra made sampling differentiable.
- The Architecture of Dreams, Part 1: How Neural Networks Learned to ImagineWorld Models, PlaNet, Dreamer.
- Finding the ManifoldWhere neural networks land: the geometry of optimisation, with a live demo.
Data & evaluation
Perspectives
- If AI Will Shape the World, the World Must Help Govern ItOpen weights, legitimacy, and who gets to participate in science.
- NeurIPS 2025: The Year We Admitted We Were LostAnd found a better map. Notes on the conference that challenged scaling orthodoxy.
- Animals, Ghosts, and ChimerasThe AI debate that will define the next decade.











