WriteSAE introduces sparse autoencoders with rank-1 matrix atoms for recurrent state updates, allowing replacement tests that outperform deletion on 92.4% of positions and a formula predicting logit changes with R²=0.98.
Bart Bussmann, Patrick Leask, and Neel Nanda
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
citation-role summary
background 1
method 1
citation-polarity summary
fields
cs.LG 5verdicts
UNVERDICTED 5representative citing papers
The paper introduces compositional interpretability as a category-theoretic framework that casts mechanistic explanations as commuting syntactic-semantic mappings optimized under faithfulness and complexity constraints derived from minimum description length.
PROPEL amortizes solver evaluation with a trained activation probe to optimize task generators toward a target solve rate, raising the share of learnable tasks from ~10% to ~20% in coding and SWE experiments.
Proposes a probabilistic framework for latent agentic substructures in DNNs using log-score utilities and log pooling, with proofs on unanimity and an application to persona emergence in LLM alignment.
Four changes to Activation Oracle training yield marginal capability gains but better practical quality, plus an open-sourced evaluation suite AObench.
citing papers explorer
-
WriteSAE: Sparse Autoencoders for Recurrent State
WriteSAE introduces sparse autoencoders with rank-1 matrix atoms for recurrent state updates, allowing replacement tests that outperform deletion on 92.4% of positions and a formula predicting logit changes with R²=0.98.