Hot AI

Cross-seed explainability using Procrustes-conditioned Joint End-to-end Top-K Sparse Autoencoders

AI·Procrustes-conditioned Joint End-to-end Top-K Sparse Autoencoders extract cross-seed universal features from independently trained BERT models to address dictionary learning misalignment in mechanistic interpretability.

arXiv cs.CL·arxiv.org·23h ago·1.5Researchinterpretability saes bert