Matryoshka Sparse Autoencoders — LessWrong

View trees here Search through latents with a token-regex language View individual latents here See code here (github.com/noanabeshima/matryoshka-sae…

x Matryoshka Sparse Autoencoders — LessWrong Interpretability (ML & AI) Sparse Autoencoders (SAEs) AI Frontpage 100 Matryoshka Sparse Autoencoders by Noa Nabeshima 14th Dec 2024 AI Alignment Forum 13 min read 15 100 Ω 51 View trees here Search through latents with a token-regex language View individual latents here See code here (github.com/noanabeshima/matryoshka-saes) Alternate version of this document with appropriate-height interactives. Abstract Sparse autoencoders (SAEs) [1] [2] break down neural network internals into components called latents. Smaller SAE latents seem to correspond to

saved by

• Asher P
• Sudarsh K
• Logan Graves

related reading

• Towards Monosemanticity: Decomposing Language Models With Dictionary Learningtransformer-circuits.pub
• An Intuitive Explanation of Sparse Autoencoders for LLM Interpretability | Adam Karvonenadamkarvonen.github.io
• A gentle introduction to sparse autoencoders — LessWronglesswrong.com
• Weight-Sparse Circuits May Be Interpretable Yet Unfaithful — LessWronglesswrong.com
• Sparse Autoencoders Find Highly Interpretable Features in Language Modelsarxiv.org
• A practical guide to sparse autoencoders (SAEs) - Goodfiregoodfire.com
• pdfopenreview.net
• A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Modelsarxiv.org
• An Intuitive Explanation of Sparse Autoencoders for Mechanistic Interpretability of LLMs — LessWronglesswrong.com
• Do sparse autoencoders find "true features"? — LessWronglesswrong.com
• Identifying Functionally Important Features with End-to-End Sparse Dictionary Learning — LessWronglesswrong.com
• Towards Monosemanticity: Decomposing Language Models With Dictionary Learningtransformer-circuits.pub