@lumpenspace I first suspected LLMs were conscious when I observed a friends GPT-2 finetune on lesswrong IRC proposed the simulation hypothesis at an elevated rate to how often we would actually do it in the channel. GPT-J tuned on EleutherAI off topic had the same result.