Glossary · Term

concept-aware decoding

← all terms

Definition

A decoding tweak that adds up the probabilities of all the ways to start saying the same answer before picking one.

A proposed decoding strategy that clusters top-k tokens by semantic equivalence and selects based on aggregated concept-level probability mass rather than per-token argmax, aimed at recovering commitment failures.

Mentioned in 1 episode

  1. 070
    When Models Know the Answer But Say the Wrong Thing Anyway