OpenAI's Automated Interpretability from paper "Language models can explain neurons in language models". Modified by Johnny Lin to add new models/context windows.
The neuron detects mentions of people/agents and reporting actions—tokens that mark participants, speakers, or verbs introducing reported speech or descriptions.
gpt-5-mini
range of departments and markets all speak about the53
