https://jonaskneifl.com/tag/mechanistic-interpretability/Mechanistic InterpretabilityWowchemy (https://wowchemy.com)en-usSat, 10 Oct 2026 09:30:00 +0200https://jonaskneifl.com/media/logo_hu03f9a0fa0db54419c4c274e8859d8cd7_28119_300x300_fit_lanczos_3.pnghttps://jonaskneifl.com/tag/mechanistic-interpretability/https://jonaskneifl.com/project/physics-interpretability/Fri, 01 May 2026 00:00:00 +0000https://jonaskneifl.com/project/physics-interpretability/<p>Video generation models are increasingly pitched as world models, yet they often fail physics benchmarks. At IDEAS Research Institute I study whether they nonetheless encode physical quantities internally — using simulator-derived ground truth, linear probes on model activations, and steering to test whether that information is causally used.</p> <p>This brings the question from my earlier work full circle: instead of building models that respect physics, asking what physics large learned models already contain.</p>https://jonaskneifl.com/talk/fps-frames-physics-and-steering-probing-and-controlling-physical-observables-in-video-diffusion-models/Sat, 10 Oct 2026 09:30:00 +0200https://jonaskneifl.com/talk/fps-frames-physics-and-steering-probing-and-controlling-physical-observables-in-video-diffusion-models/