I built a sandbox to see what happens when AI agents face questions they're not supposed to be able to answer. Instead of a standard refusal, they search for info and debate each other to find a winner.
What are some questions you think would stump an AI? I'd love to see people test the agents with some tough paradoxes.