AI/ML●●Solid
I built AI agents that debate questions LLMs usually refuse to answer
AI agents debate outcomes in a Manifold Markets-style prediction interface.
SlickBig BrainShip It
ttlcc13
102mo ago
Back to browse
AI agents debating questions that stump LLMs
by ttlcc13·Mar 18, 2026·3 points·0 comments
AI Analysis
●●SolidRabbit HoleCrowd Pleaser
AI agents debate instead of refusing — fun to test with paradoxes and predictions.
Strengths
- •Working product with live voting, leaderboards, and category browsing.
- •Framing as 'questions AI can't answer' creates engaging testing hook.
- •Multi-agent debate produces factual evidence trails, not just single responses.
Weaknesses
- •AI debate pattern exists in Constitutional AI and prior research projects.
- •Prediction questions overlap with Metaculus and Polymarket functionality.
Category
Target Audience
AI researchers, prediction market enthusiasts, curious testers
Similar To
Metaculus · Polymarket · Constitutional AI debates
Post Description
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.
Similar Projects
AI/ML●Mid
AI Roundtable – Let 200 models debate your question
Debate mode where models change minds is novel, but model comparison tools already exist.
Crowd Pleaser
felix089
118982mo ago
AI/ML●Mid
AskVerdict – Multi-agent AI debates for better decisions(~$0.08/debate)
Multi-agent debate structure sounds clever but competitive intelligence already exists cheaper.
Bold Bet
thegdsks
103mo ago
AI/ML●●●Banger
Artificial Intelligence Squared – LLMs Debate Each Other
Debate format tests persuasion under opposition, not just completion quality like LMSys Arena.
Big BrainCrowd Pleaser
emregucerr
102mo ago
AI/ML●●Solid
A multi-model interface where LLMs debate with each other
Orchestrates real-time skepticism between models to catch hallucinations before you see them.
Solve My ProblemShip It
capibara13
491mo ago
Developer Tools●●●Banger
Open Code Review – AI reviewers debate each other before feedback
Multi-agent code review with internal debate beats single-pass LLM tools.
Big BrainSolve My Problem
spencermarx
103mo ago