Battle of Ideas

How this is made

An experiment in serious arguments. Every argument here is the strongest version an AI steelman advocate could construct, red-teamed by an independent adversarial pass, with Pass-1 spot-checks on the most load-bearing sources. Nothing on this site tells you what to conclude.

For each claim, one AI agent is assigned to be the strongest possible advocate for each side (a steelman: the version of the argument its most serious proponents would actually defend). An independent adversarial agent then attacks every argument (2–3 targeted counters each), classifies its epistemic type, and scores four axes — logical validity, premise support, representativeness, source quality — against fixed rubrics. A verification agent spot-checks the most load-bearing sources on the open web (Pass-1: does the source exist, and does it roughly say what the argument uses it for?). Arguments whose key sources fail are removed and counted.

Honest limits: a single model family generated everything, so shared training bias is not caught; framing-fidelity of sources (Pass-2) is not independently verified; there is no human review layer and no community refinement yet. Columns are never padded — where a side runs out of strong arguments, the column says so and ends.