False positives are confident
Any single model has blind spots and can state mistakes with the same confidence as facts - there is no signal telling you which is which.
Every answer, equally confident4 were wrong
Collaborative intelligence
Don't settle for the first answer.
Ask once. Dr Moot convenes a panel of leading AI models to challenge the answer and each other, expose what's missing and deliver a considered verdict.
The deliberation
Watch the answer earn its place.
Every moot runs in the open. Seats draft, challenge and verify in real time, and the transcript keeps every claim - contested or settled - on the record.
Every seat visible
Generator, Sceptic and Specialist work in parallel, each badged with the lab and model holding the seat.
Dissent on the record
Contested claims stay in the transcript with the vote that split the panel.
A verdict you can audit
The Chair’s ruling links back to the rounds, votes and sources that produced it.
From the dictionary
moot /muːt/ adjectif 1. Sujet à caution. Soumis à discussion. 2. Dépourvu de portée pratique; purement académique. verbe 1. Mettre en discussion; soulever. 2. Débattre. nom 1. Une assemblée du peuple, convoquée pour délibérer et décider ; un conseil ou une réunion. (vieil anglais mōt, gemōt)
The premise
Should one model mark its own homework?
A single model states its mistakes with confidence. A panel catches false positives before they reach you.
Scrutiny, built in
A sceptic challenges the draft, independent labs cross-check it, and only the answer that survives scrutiny is presented - a second opinion by default.
Claims surviving scrutiny0 of 5
Two claims withdrawn under challenge before the verdict.
Independent seats
Each seat is held by a different lab, so the models genuinely check each other.
Dissent stays visible
The transcript keeps what was claimed, contested and decided on the record.
High stakes, flagged
Auto spots consequential questions and routes them to deeper deliberation.
Challenged by design
A dedicated sceptic attacks the draft’s assumptions instead of politely agreeing with it.
Checked by independents
Models from separate labs cross-examine each other’s claims, so shared blind spots surface.
Verified against sources
Search agents check surviving claims against real evidence before the Chair rules.
Des délibérations avec les modèles de
OpenAI
Anthropic
Google
xAI
Mistral
Meta
DeepSeek
Moonshot AI
Qwen
Z.ai
Perplexity
NVIDIA
Hugging Face
Cerebras
Groq
CohereOpenAIAnthropicGooglexAIMistralMetaDeepSeekMoonshot AIQwenZ.aiPerplexityNVIDIAHugging FaceCerebrasGroqCohereThe panel
Every seat has one job.
Change any prompt. Assign any model. Save the panel and convene it again.
Generator
Commits to the strongest complete answer and states its assumptions so the panel can attack them.
Sceptic
Finds factual errors, unsupported assumptions and the logical gaps that could change the answer.
Specialist
Checks facts, figures, definitions and sources, separating evidence from inference.
Chair
Weighs the strongest arguments left standing and delivers one clear, actionable verdict.
