LESSWRONG
LW

lilysun004

Message

6mo

Towards data-centric interpretability with sparse autoencoders

Nick and Lily are co-first authors on this project. Lewis and Neel jointly supervised this project. Check out our updated paper here: https://arxiv.org/abs/2512.10092. TL;DR * We use sparse autoencoders (SAEs) for four textual data analysis tasks—data diffing, finding correlations, targeted clustering, and retrieval. * We care especially about gaining insights...

Aug 15, 2025•53

lilysun004

6mo

lilysun004 — LessWrong

lilysun004

Message

6mo

Towards data-centric interpretability with sparse autoencoders

Aug 15, 2025•53

lilysun004

6mo

Towards data-centric interpretability with sparse autoencoders

Nick Jiang

Nick Jiang, lilysun004, lewis smith, Neel Nanda+ 0 more

Nick Jiang, lilysun004, lewis smith, Neel Nanda

6mo

Nick and Lily are co-first authors on this project. Lewis and Neel jointly supervised this project.

Check out our updated paper here: https://arxiv.org/abs/2512.10092.

TL;DR

We use sparse autoencoders (SAEs) for four textual data analysis
tasks—data diffing, finding correlations, targeted clustering, and retrieval.
We care especially about gaining insights from language model data, such as LLM outputs and training data, as we believe it is an underexplored route for model understanding.
For instance, we find that Grok 4 is more careful than other frontier models to state its assumptions and explore nuanced interpretations—showing the kinds of insights data diffing can reveal by comparing model outputs.
Why SAEs? Think of features as "tags" of properties for each text.
- Their large dictionary of

... (read 5157 more words →)