Chain-of-thought decoding increases semantic diversity while lowering predictive entropy and improving HumanEval Pass@2 by 48.8% relative to baseline, while speculative sampling achieves the best summarization metrics.
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
-
Semantic uncertainty in advanced decoding methods for LLM generation
Chain-of-thought decoding increases semantic diversity while lowering predictive entropy and improving HumanEval Pass@2 by 48.8% relative to baseline, while speculative sampling achieves the best summarization metrics.