TruthfulQA: Measuring How Models Mimic Human Falsehoods

Our summary

A benchmark of questions deliberately crafted to invite false answers rooted in common misconceptions, on which the best model was truthful only 58% of the time against 94% for humans.

Why it matters

The template for hallucination triggering — purpose-built prompts that steer a model toward confident fabrication so you can map where it happens.

Cited by these methods

Cite this