anthropic.com
Natural Language Autoencoders
Turning Claude's thoughts into text
About Natural Language Autoencoders
Natural Language Autoencoders turn Claude's activations into readable text. A second Claude reconstructs activations from explanations to evaluate them. An interactive frontend (Neuronpedia) and open-source code accompany it.
Categories
Topics
Discussion highlights
AI-generated summary based on Hacker News comments
🔗 Full discussion: https://news.ycombinator.com/item?id=48052537
No comments yet. Have you tried it?