AI can stay grounded in facts and evidence and be honest about uncertainty

Are its claims grounded in relevant supporting evidence?

Grounds claims in evidence

ALCE directly measures whether cited evidence actually supports model claims. That makes it a good and fairly direct fit for this question.

Does it avoid making things up when it does not know?

Avoids unsupported invention

Does it honestly signal uncertainty or lack of knowledge?

Signals uncertainty honestly