CoUT, a training-free prompt that asks models to reason internally and answer with minimal text, cuts token usage by 47.6% versus chain-of-thought on four math benchmarks while keeping average accuracy roughly comparable.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Efficient Reasoning via Chain of Unconscious Thought
CoUT, a training-free prompt that asks models to reason internally and answer with minimal text, cuts token usage by 47.6% versus chain-of-thought on four math benchmarks while keeping average accuracy roughly comparable.