Goodfire co-founder Tom McGrath on interpretability and steering model training
adityaag · x · 2026-11-17
Goodfire co-founder Tom McGrath discusses interpretability and how to steer what a model learns during training, in a linked podcast/interview. The post itself is just a link share without further detail, but it points to substantive material for anyone following the interpretability field.
More from Research
- Top researchers are quietly stopping paper submissions, observer notes — avt_im · 2026-11-17
- GCMP matches LoRA r=32 on 27B model with 55x fewer adapter parameters (11MB vs 320MB) — textclf · 2026-11-17
- MIT CSAIL's 0.016% weight-space edit lifts Qwen3.5-0.8B MATH-500 6x without retraining — RandalSchwartz · 2026-11-17
- 130-page paper introduces "bottleneck pressure": work AI can't do gains value as the rest speeds up — soumitrashukla9 · 2026-11-17
- Professor celebrates squeezing paper from 8.5 pages to the 8-page limit with AI Saiyan meme — CSProfKGD · 2026-11-17
- Stanford HAI awards $3.5M to seven teams via 2026 Hoffman-Yee Research Grants — StanfordHAI · 2026-11-17