Paper
Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet
Used Sparse Autoencoders (SAEs) to "de-mix" the activations of Claude 3 Sonnet, identifying millions of individual "features" like "Golden Gate Bridge" or "G...
Related papers
21
-
Anthropic
-
MATS
-
Theorem
-
UK AISI
-
Alignment Research Center
-
Anthropic
-
Truthful AI
-
University of California, Berkeley
-
Warsaw University of Technology
-
Center for AI Safety
-
Center for AI Standards and Innovation
-
Co-affiliated with Carnegie Mellon University
-
UK Artificial Intelligence Security Institute
Looking for more?
Open the arXiv / PDF link for the full paper. Abstract and authors appear here when enrichment has run.