8 papers
SynthSAEBench: Evaluating Sparse Autoencoders on Scalable Realistic Synthetic Data
David Chanin, Adrià Garriga-Alonso
Improving Sparse Autoencoders (SAEs) requires benchmarks that can precisely validate architectural innovations. Current LLM-based SAE benchmarks are too noisy to differentiate arch…
Sparse but Wrong: Incorrect L0 Leads to Incorrect Features in Sparse Autoencoders
David Chanin, Adrià Garriga-Alonso
Sparse Autoencoders (SAEs) extract features from LLM internal activations, meant to correspond to interpretable concepts. A core SAE training hyperparameter is L0: how many SAE fea…
Biases in the Blind Spot: Detecting What LLMs Fail to Mention
Iván Arcuschin, David Chanin, Adrià Garriga-Alonso +1
Large Language Models (LLMs) often provide chain-of-thought (CoT) reasoning traces that appear plausible, but may hide internal biases. We call these unverbalized biases. Monitorin…
Are Sparse Autoencoder Benchmarks Reliable?
David Chanin
Sparse autoencoders (SAEs) are a core interpretability tool for large language models, and progress on SAE architectures depends on benchmarks that reliably distinguish better SAEs…
A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders
David Chanin, James Wilken-Smith, Tomáš Dulka +3
Sparse Autoencoders (SAEs) aim to decompose the activation space of large language models (LLMs) into human-interpretable latent directions or features. As we increase the number o…
Feature Hedging: Correlated Features Break Narrow Sparse Autoencoders
David Chanin, Tomáš Dulka, Adrià Garriga-Alonso
It is assumed that sparse autoencoders (SAEs) decompose polysemantic activations into interpretable linear directions, as long as the activations are composed of sparse linear comb…