- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper proposes ConfSFT, which trains reasoning models to predict confidence at intermediate reasoning states using targets derived from their own answer-token probabilities. Training supervises only confidence labels, and inference remains unchanged. Across four model families and five benchmarks, the method reduces average generated tokens by approximately 10–19%, with generally small average accuracy changes. Ablations examine supervision targets and decision-point selection.
Critical strengths
-
Practical simplicity: Confidence targets require neither gold answers nor external judges, and deployment requires no confidence queries or stopping mechanism. This makes the approach potentially useful with little additional inference complexity.
-
Broad empirical support: Token reductions occur across multiple model families and task domains. The shuffled-label ablation provides meaningful evidence that the correspondence between reasoning states and supervision targets matters.
Critical weaknesses
-
Accuracy preservation is overstated: Table 5 reports negative nominal 95% accuracy intervals for several settings, including Qwen on LiveCodeBench (−1.9 ± 1.3 percentage points) and GPT-OSS on LiveCodeBench (−2.9 ± 2.3). These qualify the central “matched accuracy” claim; nonsignificant differences elsewhere do not establish equivalence.
-
Evaluation and reproducibility remain insufficiently resolved: The confidence-interval construction and exact checkpoint-selection criteria are unclear. Qwen’s identical reported GSM8K pass@1 and pass@8 values warrant verification.
Questions for authors
-
How were the paired confidence intervals constructed, and how should the accuracy-preservation claim be reconciled with the negative intervals in Table 5?
-
Are Qwen’s identical GSM8K pass@1 and pass@8 values correct? Please verify the stated independent sampling procedure using per-question correct-completion counts.
-
What exact checkpoint-selection rules were used? In particular, was A&Z selected on AIME2025, as Appendix F.2 states, despite the earlier statement excluding test-set model selection?