12 July 2026Reading notes · 8 min Reading CausalPFN: causal effects as in-context predictionA NeurIPS 2025 paper trains one transformer on simulated, identifiable causal worlds; new observational data goes in as context, treatment-effect posteriors come out. What holds it up, and where it breaks. ↗
5 July 2026Continual learning · 9 min The geometry of forgetting in continual learningHow learning a new task damages an old one—and how interference geometry suggests a structural controller. ↗
28 June 2026Electrode manufacturing · 7 min Solving the electrode calendering U-shape puzzleWhy particle reorientation reverses the monotone prediction made by porosity-only thermal models. ↗