What does agreement between models actually justify?
What I think right now
Less than it looks like. Models that share training data agree for shared reasons, so consensus has to be discounted by where the inputs came from. The Polymorphic Decision Protocol tests confidence under controlled pressure and leaves the weights, the stakes and the exit with a person.
What would change my mind
Evidence that independently trained models fail independently.
Loose ends
- How to price disagreement.
- When a panel of models is worse than one well-checked answer.
The trail
- Essay Distance from Gray A commitment-specific account of what remains unestablished, and what contracts and measurement can show without pretending to prove truth.
- Research Polymorphic Decision Protocol ↗ A specified decision procedure separates evidence, pressure testing, human judgment, and prospective evaluation.
Think I have this wrong? That is the most useful thing you could tell me.
Disagree with me