Chris Potts' IPAM talk on interpretability and subliminal learning now available
ChrisGPotts · x · 2026-09-06
Stanford NLP shared Chris Potts' talk at the IPAM Foundations of Interp workshop (UCLA), covering interpretability, representations, data, and subliminal learning — praised as an excellent explanation of the main ideas of the paper. Video link is public.
More from Research
- PyTorch gotcha: scalar store/load without a debug_barrier leads to a very bad time — typedfemale · 2026-09-06
- General Intelligence Index: psychometric g-factor over 59 benchmarks and 267 models — wyatt400 · 2026-09-06
- ECCV 2026 to host Geometric Intelligence workshop on vision for scientific discovery — tolga_birdal · 2026-09-06
- SkillGLoW: agents that remember procedures, not tasks, gain 17.2 points with 3.6x smaller memory — rohanpaul_ai · 2026-09-06
- John Langford shares discussion on why and what for world models, with Information Bottleneck folks — JohnCLangford · 2026-09-06
- PaperLens MCP Verifies Papers Are Actually Implemented in Code, Now Open Source — CraftyScore4348 · 2026-09-06