Partial Reasoning in Language Models: Search and Refinement Guided by Uncertainty
Workshop Paper ·
January 2026
· 1 min read
We propose partial reasoning in language models guided by uncertainty, using search and refinement to selectively truncate chain-of-thought while preserving performance.
We propose a partial reasoning framework for language models where search and refinement are guided by uncertainty estimates, selectively truncating chain-of-thought reasoning while preserving downstream task performance.