2 papers
cs.CL2026
Turn-Averaged SAEs for Feature Discovery and Long-Context Attribution
Kevin Der, Harish Kamath, Ben Thompson
Sparse autoencoders (SAEs) have become a useful tool for extracting interpretable features in language models. However, standard SAE architectures operate on individual token activ…
cs.LG2025
Distillation Robustifies Unlearning
Bruce W. Lee, Addie Foote, Alex Infanger +6
Current LLM unlearning methods are not robust. A few steps of finetuning can revert their effects. We begin by showing that this is true even for an idealized form of unlearning: t…