Interpreting the Black Box: Natural Language Autoencoders Translate LLM Activations into Readable…

Towards AI
Generative AI

Teaching Language Models to Describe Their Own Internal State (Confabulation Included)