New Approach to Block-Sparse Featurizers
burny_tech · x · 2026-07-09
A post shares a research update from Goodfire proposing Block-Sparse Featurizers (BSFs), which discover concepts in model activations using multi-dimensional "blocks" rather than individual directions. The forwarded content also mentions that their AI neuroscientist, "silico," can automatically train BSFs for in-progress or open-source models, with early access now available.
Related event: Goodfire Releases Block-Sparse Featurizers for Model Interpretability(3 posts)→
More from Research
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- Terence Tao's SAIR Foundation revamps Mathathon with $1.5M compute per team — _sathvikr · 2026-10-27
- Mathathon 2.0 launches with Terence Tao's SAIR, $1.5M in compute for teams — _sathvikr · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27
- PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction — kwangmoo_yi · 2026-10-27
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27