Attribution-based parameter decomposition — LessWrong

This is a linkpost for Apollo Research's new interpretability paper:  …

x Attribution-based parameter decomposition — LessWrong Interpretability (ML & AI) Apollo Research (org) AI Frontpage 2025 Top Fifty: 14 % 109 Attribution-based parameter decomposition by Lucius Bushnaq , Dan Braun , StefanHex , jake_mendel , Lee Sharkey 25th Jan 2025 AI Alignment Forum Linkpost for publications.apolloresearch.ai 5 min read 21 109 Ω 45 This is a linkpost for Apollo Research's new interpretability paper: " Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition ". We introduce a new method for directly decomp

saved by

• Yudhister Joel Kumar

related reading

• Stochastic Parameter Decompositionarxiv.org
• An Extremely Opinionated Annotated List of My Favourite Mechanistic Interpretability Papers v2 — AI Alignment Forumalignmentforum.org
• Interpreting Language Model Parametersgoodfire.ai
• AlgZoo: uninterpreted models with fewer than 1,500 parameters — LessWronglesswrong.com
• Weight-Sparse Circuits May Be Interpretable Yet Unfaithful — LessWronglesswrong.com
• Towards Scalable Parameter Decompositiongoodfire.ai
• [2501.14926] Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decompositionarxiv.org
• Towards Scalable Parameter Decompositiongoodfire.ai
• Towards Monosemanticity: Decomposing Language Models With Dictionary Learningtransformer-circuits.pub
• A Mathematical Framework for Transformer Circuitstransformer-circuits.pub
• [2506.20790] Stochastic Parameter Decompositionarxiv.org
• An Extremely Opinionated Annotated List of My Favourite Mechanistic Interpretability Papers v2 — AI Alignment Forumalignmentforum.org