Mechanistic Interpretability Hackathon 5 runs Nov 20-22, theme Autointerp Evals
burny_tech · x · 2026-11-19
The 5th Mechanistic Interpretability Hackathon runs Nov 20-22 (48 hours) on the Mechinterp Discord, with submissions to itch.io. Official theme: Autointerp Evals — building ways to automatically score interpretability explanations, similar to OpenAI's autointerp scoring explanations by how well they predict activations. Other suggested directions: testing Nick Cammarata's 'neurons are already clean' claim, steering neurons with clear meanings, or exploring J-lens. Toy-size projects welcome; kickoff, discussion, and closing sessions daily at 5pm UTC.
More from Companies & People
- OpenAI's own whisper repo shows a commit co-authored by Claude, complete with screenshot — Arindam_1729 · 2026-11-19
- Uber, SpaceX and Nvidia engineers pack first AI Software Factory summit PortCon — Pavan_Belagatti · 2026-11-19
- Microsoft kills To Do app, tells users to move to Planner — LuoSays · 2026-11-19
- Report: Jensen Huang's daughter clashed with Nvidia execs as marketing director — Kyrannio · 2026-11-19
- DeepMind interpretability researcher's team pivots fully to frontier AI safety, hiring in Seattle — _beenkim · 2026-11-19
- NVIDIA employee says Jensen was 'perhaps the only one who could' lead that fight, lambasts board — Kyrannio · 2026-11-19