mlsanityData to 13 Oct 2026

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

arXiv 2410.12555 ↗First posted 16 Oct 2024cs.LGcited by 2

On the map: Mechanistic interpretability → SAE features and their applications

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.

Key facts. Cited by 2 papers in the map (data through 13 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.

Builds on

4 of 10 references.

Closest ancestors · 1

Same direction, newest first

  1. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Models15 Sep

Other references · 3

  1. 2016Linear Algebraic Structure of Word Senses, with Applications to Polysemycited by 165
  2. 2024Identifying Functionally Important Features with End-to-End Sparse Dictionary Learningcited by 66
  3. 2024Mathematical Models of Computation in Superpositioncited by 35

Built on by

Cited by year

Citing papers per year: 2024: 2.

2 papers.

  1. 2024Decomposing The Dark Matter of Sparse Autoencoderscited by 45
  2. 2024Features that Make a Difference: Leveraging Gradients for Improved Dictionary Learningcited by 5

Similar, not cited

Close in wording, no citation either way

  1. 2025SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability
  2. 2025Tokenized SAEs: Disentangling SAE Reconstructions
  3. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
  4. 2026Are Sparse Autoencoder Benchmarks Reliable?
  5. 2025AbsTopK: Rethinking Sparse Autoencoders For Bidirectional Features
  6. 2026When Pruning Meets Interpretability: Preserving Sparse Autoencoder Robustness in LLMs
  7. 2026When Can We Trust the Sparse Lens? A Certification Framework for SAE Faithfulness
  8. 2024Improving Dictionary Learning with Gated Sparse Autoencoders