Definition
How confidently or hesitantly an AI's writing sounds.
A measurable property of LLM outputs scored via balanced lexicons of assertive and hedging terms, used to mediate the effect of capability on attack success in the Capability Paradox paper.