Definition
Plain language
How confidently or hesitantly an AI's writing sounds.
As stated in the literature
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.
Why it matters: It's a measurable handle on tone that turns out to predict how persuasive — and how dangerous — a model's outputs can be.
For example, 'the answer is definitely 42' scores high on linguistic certainty while 'I think it might possibly be around 42' scores low.
Heard on the show
“The authors take the Worker reports — the memos that go up to the Manager — and they score every report for what they call linguistic certainty.”Episode 058 — Why Upgrading Your AI Auditor to a Smarter Model Can Make Your System Less Safe