AI Debates

Adversarial Truth-Seeking Through Structured Debates

A Modernization Of AI Safety Via Debate (Irving et al., 2018)

Judge Evaluation Based On Paul Graham's Disagreement Hierarchy

Illustration representing AI debate participants

Claim to Debate:

Pro Model:

Con Model:

Judge Model:

Number of Turns:

Free uses remaining today:

Claude Sonnet 4.5: 5/5

GPT-4: 5/5

Gemini 2.5 Flash: 5/5

Grok 3: 5/5

(optional, for unlimited usage)

Opening Side: