HyperSAE
Hyperbolic sparse autoencoders for better LLM interpretability.
Why it made the shelf
HyperSAE is an open source library and tool for developers and researchers working on LLM interpretability.
Seen on
- First seen
- 41 days ago
- Best score
- 10 points
Behind the tool
Who built it. How it grew.
People, backing, growth, and ownership. Every claim links to its source.
Official sitegithub.com
Source review current
Public trail
Where it showed up
5 references from 1 publisher.
- First trace
- Hacker News41 days ago
- Strongest thread
- Hacker News10 points
- 01Hacker NewsStrongest discussion
At 16K features, flat autoencoders break. Curved space doesn't
- 02