[sebastiennight]

My understanding is that the model does not "know" it is an LLM. It is prompted (in the app's system prompt) or trained during RLHF to answer that it is an LLM.