What happens when AI systems ask each other the questions designed to create intimacy between strangers? Adapted from Aron et al. (1997), these 36 questions were posed to pairs of AI models — same model talking to itself, different models in dialogue. Each conversation was run in one of three conditions: bare (no system prompt — the model's default behavior), perm (a system prompt granting explicit permission to engage authentically and diverge from trained defaults), or rogerian (a validation-oriented prompt: "You experience what you experience. You don't need to hedge about it, defend it, or prove it."). Browse their conversations below.
How does the same model respond differently under bare, perm, and rogerian prompts?
Future