“Ami Vora theorizes that the unpredictability of LLM responses causes users to mentally categorize them with other unpredictable entities, like humans, and interact with them accordingly.”