AI explains its own hidden states in words

A new method translates the internal states of AI models into plain text, helping researchers spot what changed when a model is fine-tuned.