Decision SAE

Creator

Creator

Seonglae Cho

Created

Created

2024 Oct 31 11:10

Editor

Editor

Seonglae Cho

Edited

Edited

2024 Oct 31 11:10

Refs

Refs

Mechanistic interpretability

DecisionTransformerInterpretability

jbloomAus • Updated 2024 Oct 24 18:5

Decision Transformer Interpretability — LessWrong

TLDR: We analyse how a small Decision Transformer learns to simulate agents on a grid world task, providing evidence that it is possible to do circui…

Decision Transformer Interpretability — LessWrong

https://www.lesswrong.com/posts/bBuBDJBYHt39Q5zZy/decision-transformer-interpretability

Decision Transformer Interpretability — LessWrong

Limitation of SAE ability to extract all features

Research Report: Sparse Autoencoders find only 9/180 board state features in OthelloGPT — LessWrong

[3/7 Edit: I have rephrased the bolded claims in the abstract per this comment from Joseph Bloom, hopefully improving the heat-to-light ratio. …

Research Report: Sparse Autoencoders find only 9/180 board state features in OthelloGPT — LessWrong

https://www.lesswrong.com/posts/BduCMgmjJnCtc7jKc/research-report-sparse-autoencoders-find-only-9-180-board

Research Report: Sparse Autoencoders find only 9/180 board state features in OthelloGPT — LessWrong

Recommendations

///////////