# Investigating Sensitive Directions in GPT-2: An Improved Baseline and Comparative Analysis of SAEs

arXiv: https://arxiv.org/abs/2410.12555 · HTML: https://mlsanity.com/paper/2410.12555/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in SAE features and their applications, one of 13 directions in Mechanistic interpretability.
- Builds on 4 of its 10 references that are in the map.
- First posted on arXiv 16 Oct 2024.

## Placed in

- Mechanistic interpretability / [SAE features and their applications](https://mlsanity.com/mechinterp/sae-features-and-their-applications/)

## Builds on (most cited first, 4 of 4)

- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Linear Algebraic Structure of Word Senses, with Applications to Polysemy](https://mlsanity.com/paper/1601.03764/) (2016)
- [Identifying Functionally Important Features with End-to-End Sparse Dictionary Learning](https://mlsanity.com/paper/2405.12241/) (2024)
- [Mathematical Models of Computation in Superposition](https://mlsanity.com/paper/2408.05451/) (2024)

## Built on by (most cited first, 2 of 2)

- [Decomposing The Dark Matter of Sparse Autoencoders](https://mlsanity.com/paper/2410.14670/) (2024)
- [Features that Make a Difference: Leveraging Gradients for Improved Dictionary Learning](https://mlsanity.com/paper/2411.10397/) (2024)

## Abstract (arXiv)

Sensitive directions experiments attempt to understand the computational features of Language Models (LMs) by measuring how much the next token prediction probabilities change by perturbing activations along specific directions. We extend the sensitive directions work by introducing an improved baseline for perturbation directions. We demonstrate that KL divergence for Sparse Autoencoder (SAE) reconstruction errors are no longer pathologically high compared to the improved baseline. We also show that feature directions uncovered by SAEs have varying impacts on model outputs depending on the SAE's sparsity, with lower L0 SAE feature directions exerting a greater influence. Additionally, we find that end-to-end SAE features do not exhibit stronger effects on model outputs compared to traditional SAEs.