[KongchangAI]
ConceptExplicit Confidence Targets

Explicit confidence target

A mechanism that lets users specify a target confidence value p in the prompt, according to which the model decides whether to answer or abstain, with the goal that the accuracy of answered questions reaches at least p, internalizing confidence control into the model policy.

Timeline (last 90 days)

Oct 27

An explicit confidence target means the user specifies a target value p in the prompt, and the model accordingly directly chooses whether to answer or abstain, with the goal that the accuracy on answered questions reaches at least p.

Unverified50%
Oct 27

Whether explicit confidence targets can remain stable in more complex multi-turn interaction or tool-calling scenarios is currently unclear.

Unverified50%

All Facts (2)

Source Articles