- Rating
- 2
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper studies a limitation of majority voting in inference-time scaling: an incorrect answer may receive more votes than the correct answer because many different reasoning paths lead to it. The authors analyze this problem using a KL-regularized policy model and show that answer probability depends on both reward and the reference policy's probability mass for that answer. To address this issue, the paper proposes SELFOCUS, a training-free answer selection method. It clusters semantically similar intermediate reasoning steps across sampled CoTs, assigns lower scores to frequently occurring steps, and combines these scores with answer frequency to select the final answer. Experiments across four model families and seven benchmarks show improvements over majority voting
Critical strengths
-
Consistent empirical improvements. SELFOCUS outperforms majority voting and Semantic Self-Consistency (SSC) in 15 of 17 settings, with particularly consistent gains on TheoremQA across three model families.
-
A practical, training-free approach to answer selection. SELFOCUS uses intermediate-step semantic recurrence without requiring additional training or token log probabilities. This provides an alternative to probability-based selection methods. The paper also includes useful ablations and controlled experiments to examine the proposed method.
Critical weaknesses
-
The advantage over existing selection methods is not sufficiently compelling
Inference-time selection has been extensively studied, and SELFOCUS mainly extends semantic agreement from complete CoTs to intermediate reasoning steps. Although the results are encouraging, the improvement over SSC averages only 1.68 percentage points. Moreover, the full procedure involves step segmentation, semantic clustering, and NLI classification, but its computational cost is not compared with SSC. The small positional discounts in some settings (e.g., η = 0.1 in Table 11) also raise the question of whether a much simpler early-step selection method could achieve similar results.
-
Insufficient evidence for the claimed path-volume correction
The paper argues that incorrect answers can accumulate votes from diverse reasoning paths, while correct answers tend to have more consistent paths. However, correct answers can also be reached through different valid approaches, and systematic errors can produce highly consistent incorrect reasoning. Although Figure 3 shows a correlation between semantic recurrence and correctness, the paper does not directly establish that the proposed path-volume bias explains the majority-voting errors corrected by SELFOCUS. The controlled experiments in Appendix D.2 are useful, but they do not establish that path-volume correction explains the accuracy gains on the evaluated reasoning benchmarks.
Questions for authors
-
How does SELFOCUS compare with a simpler selector using only the first few reasoning steps, in terms of both accuracy and computational cost? Can the authors also report its computational cost relative to SSC?
-
For cases where SELFOCUS corrects majority voting, are the incorrect majority answers actually supported by more diverse reasoning paths? How does the method behave when correct answers are reached through different valid approaches?
4 comments
2. Computational Cost Relative to SSC and Scaling Potential
SSC and SELFOCUS use exactly the same 64 generated CoTs, so their generation costs are identical. We compare only post-processing: SSC embeds complete CoTs and constructs similarity-based voting weights, whereas SELFOCUS clusters reasoning steps and computes trajectory offsets.
The first table reports post-processing GPU-hours per question at , following the convention used in our paper.
| Model | Dataset | SSC GPU-h/question | SELFOCUS GPU-h/question |
|---|---|---|---|
| Gemma2-9B | GPQA | 0.00191 | 0.01350 |
| MATH500 | 0.00173 | 0.00920 | |
| TheoremQA | 0.00169 | 0.00940 | |
| Llama3.3-70B | GPQA | 0.00288 | 0.02000 |
| MATH500 | 0.00196 | 0.01590 | |
| TheoremQA | 0.00201 | 0.01620 | |
| AIME25 | 0.00262 | 0.04900 | |
| AIME26 | 0.00260 | 0.04430 | |
| Qwen3-32B | GPQA | 0.00282 | 0.02850 |
| MATH500 | 0.00225 | 0.01320 | |
| TheoremQA | 0.00217 | 0.00770 | |
| AIME25 | 0.02377 | 0.02470 | |
| AIME26 | 0.01716 | 0.01800 | |
| GPT-OSS-120B | AIME25 | 0.11044 | 0.13790 |
| AIME26 | 0.08542 | 0.11910 | |
| BRUMO25 | 0.00488 | 0.04540 | |
| HMMT25 | 0.00753 | 0.06590 |
Because accuracy gain is measured over a complete dataset, we multiply the per-question cost by the number of questions. Let be the accuracy of method , let be the accuracy of Vote, and let be the total post-processing GPU-hours of method over the complete dataset. We define compute efficiency as
Let denote the efficiency of SELFOCUS and the efficiency of SSC. We compare the two methods using
A positive indicates that SELFOCUS obtains more accuracy improvement per absolute GPU-hour.
| Model | Dataset | SSC total GPU-h | SELFOCUS total GPU-h | SSC gain (pp) | SELFOCUS gain (pp) | SELFOCUS - SSC (pp/GPU-h) |
|---|---|---|---|---|---|---|
| Gemma2-9B | GPQA | 0.379 | 2.673 | +0.51 | +2.27 | 0.849 - 1.347 = -0.498 |
| MATH500 | 0.864 | 4.600 | +0.70 | +0.80 | 0.174 - 0.811 = -0.637 | |
| TheoremQA | 1.264 | 7.022 | +0.80 | +2.61 | 0.372 - 0.633 = -0.261 | |
| Llama3.3-70B | GPQA | 0.570 | 3.960 | +0.51 | +1.01 | 0.255 - 0.895 = -0.639 |
| MATH500 | 0.982 | 7.950 | +0.20 | +0.90 | 0.113 - 0.204 = -0.091 | |
| TheoremQA | 1.499 | 12.101 | +1.20 | +4.15 | 0.343 - 0.801 = -0.458 | |
| AIME25 | 0.079 | 1.470 | 0.00 | +3.33 | 2.265 - 0.000 = +2.265 | |
| AIME26 | 0.078 | 1.329 | 0.00 | +1.66 | 1.249 - 0.000 = +1.249 | |
| Qwen3-32B | GPQA | 0.558 | 5.643 | +1.01 | +3.28 | 0.581 - 1.810 = -1.228 |
| MATH500 | 1.123 | 6.600 | +0.20 | +0.30 | 0.046 - 0.178 = -0.133 | |
| TheoremQA | 1.624 | 5.752 | +1.61 | +4.22 | 0.734 - 0.991 = -0.257 | |
| AIME25 | 0.713 | 0.741 | 0.00 | 0.00 | 0.000 - 0.000 = 0.000 | |
| AIME26 | 0.515 | 0.540 | 0.00 | +2.50 | 4.630 - 0.000 = +4.630 | |
| GPT-OSS-120B | AIME25 | 3.313 | 4.137 | 0.00 | +3.33 | 0.805 - 0.000 = +0.805 |
| AIME26 | 2.563 | 3.573 | 0.00 | +1.67 | 0.467 - 0.000 = +0.467 | |
| BRUMO25 | 0.146 | 1.362 | 0.00 | -3.33 | -2.445 - 0.000 = -2.445 | |
| HMMT25 | 0.226 | 1.977 | -1.67 | +5.00 | 2.529 - (-7.391) = +9.920 |
SSC is cheaper in every setting and is therefore more cost-efficient on smaller models and conventional benchmarks, although SELFOCUS achieves larger raw accuracy gains in most cases.
This pattern changes on high-difficulty tasks. SSC matches Vote in seven of eight settings and declines on GPT-OSS HMMT25, whereas SELFOCUS improves accuracy in six. SELFOCUS therefore provides a higher return per GPU-hour in six settings, ties in one, and is lower only on BRUMO25.
Because SSC encodes complete CoTs, its cost increases with trajectory length. On Qwen AIME25/26, its cost is already close to SELFOCUS ( vs. and vs. GPU-hours per question), while providing no accuracy gain over Vote.
We also observe that SSC increasingly approaches ordinary Vote as model scale, task difficulty, and trajectory length increase. SSC compresses each complete CoT into one global embedding and weights votes using similarities between complete trajectories. Long trajectories leading to different answers can share substantial problem context, reasoning templates, and intermediate computations. This shared content can dilute local differences relevant to answer selection, making the resulting similarity weights less discriminative and causing weighted voting to approach ordinary Vote.
Avoiding a single whole-sequence representation is an explicit design motivation of SELFOCUS. It decomposes each CoT into reasoning steps and measures the recurrence and deviation of local semantic states across trajectories. Thus, even when complete CoTs share substantial content, SELFOCUS can preserve differences in their local reasoning structures. Its additional post-processing is intended to avoid excessive compression of long-horizon reasoning into one global embedding.
to be continued in next comment...
part of comment 2 and 3. Does Path Deviation Actually Drive Answer Correction?
We do not claim a universal scaling law or that SELFOCUS is always more cost-efficient than SSC. More precisely, SSC is a low-cost global semantic selector with limited accuracy gains, whereas SELFOCUS exchanges additional step-level processing for stronger answer discrimination. The current results suggest that this design may have greater potential as model scale, generated text volume, and task difficulty increase.
3. Does Path Deviation Actually Drive Answer Correction?
We interpret the reviewer's question as raising two concerns. First, because the final SELFOCUS score combines cumulative offset, positional weighting, and answer-frequency normalization, some Vote errors might be corrected by these additional operations rather than by a lower correct-answer offset. Second, correct answers supported by heterogeneous reasoning paths may themselves receive high offsets and be penalized.
Are Vote errors corrected because of lower offsets?
To isolate the offset signal, we set , removing positional weighting, and directly compare the correct answer with the most-voted incorrect answer across all instances in which SELFOCUS corrects Vote.
| Statistic | Correct answer | Most-voted incorrect answer |
|---|---|---|
| Mean cumulative offset | 0.9868 | 1.5145 |
| Paired offset difference | -- | +0.5276 |
| 95% cluster-bootstrap CI | -- | [0.4343, 0.6259] |
| Proportion with higher incorrect-answer offset | -- | 91.3% |
In of corrected instances, the most-voted incorrect answer has a higher cumulative offset. Moreover, answer-frequency normalization favors the incorrect majority: because SELFOCUS minimizes mean offset divided by answer frequency, the larger vote count lowers the incorrect answer's score. The correct answer must therefore have a sufficiently lower offset to overcome, rather than benefit from, the frequency term.
A small fraction of instances do not satisfy this ordering, so we do not claim that every correction is entirely offset-driven. Nevertheless, the paired means, confidence interval, and directional agreement show that lower step-level cumulative offset is the primary signal in most corrected cases.
What happens when correct reasoning paths are heterogeneous?
A correct answer reached through different reasoning approaches may have lower semantic recurrence and therefore a higher offset. We operationalize this heterogeneity using the mean undiscounted step offset of trajectories supporting the correct answer. Among instances that Vote originally answers correctly, we construct tertiles using fixed boundaries of and from the analysis, and apply the same boundaries at .
| Heterogeneity | Correct-answer mean step offset | Retention at | Retention at |
|---|---|---|---|
| Low | [0.0234, 0.5002] | 99.21% | 99.15% |
| Medium | (0.5002, 0.9744] | 98.20% | 98.36% |
| High | (0.9744, 3.5730] | 94.82% | 96.51% |
Without positional weighting, the risk of overturning an originally correct Vote prediction increases with correct-answer path heterogeneity. Thus, heterogeneous correct reasoning can indeed be adversely affected by the offset score.
Early-step positional weighting partially mitigates this failure mode. In the high-heterogeneity group, correct-answer retention increases from to . In the fixed-coefficient diagnostic, positional weighting slightly reduces beneficial corrections but produces a larger reduction in harmful reversals. Its main empirical role is therefore to protect answers that Vote already selects correctly, particularly when they are supported by heterogeneous reasoning paths.
Under the final protocol with label-free estimated , SELFOCUS preserves of predictions that Vote originally answers correctly, with an overall correction rate of and a regression rate of only . Together, these results indicate that cumulative offset primarily identifies incorrect majority answers with high path deviation, while positional weighting reduces the damage to heterogeneous correct reasoning. Although the mechanism behind this protective effect requires further study, both the fixed-coefficient transitions and the heterogeneity-stratified results support this functional distinction.
If our interpretation correctly captures the reviewer's concerns, we will incorporate these analyses into the revised manuscript. Specifically, we will add the main findings and discussion to the main text and provide the complete experimental definitions, per-setting results, and computational-cost analysis in the appendix.
Thank you for your response. My concerns remain. I have decided to retain my score.
### 1. Comparison with Prefix-Only Selectors
We thank the reviewer for suggesting a comparison with simpler selectors that use only the first few reasoning steps. We evaluate two representative prefix-based methods:
We replay both methods using the same candidate trajectories as Vote and SELFOCUS. PoLR uses the first 256 tokens of each trajectory, whereas Prefix-Confidence uses the first 32 tokens. Compute is reported as GPU-hours per question at N=64, consistent with the usage convention in our paper. Each entry below is formatted as accuracy / GPU-hours per question.
The results confirm that the first few reasoning steps contain signals that can be used to filter trajectories and substantially reduce inference cost. PoLR uses approximately 41.5%--86.9% of the per-question compute required by full Vote, but its macro-average accuracy decreases from 61.76% to 59.21%. Prefix-Confidence is substantially cheaper, using approximately 1.7%--23.6% of the Vote compute, but its macro-average accuracy decreases to 50.54%. In contrast, SELFOCUS achieves a macro-average accuracy of 63.74%.
These results distinguish reducing the generation budget from improving final answer selection. Prefix signals can identify trajectories that are inexpensive to continue, but on difficult problems the first few reasoning steps do not contain sufficient information to distinguish final answers reliably. SELFOCUS instead uses the semantic recurrence and deviation structures accumulated throughout complete CoTs, retaining later reasoning information that prefix-only selectors cannot use.
Our results therefore do not suggest that prefixes contain no quality signal. Rather, under our evaluation settings, prefix information is primarily useful for reducing the inference budget. Using the step-level structure of complete CoTs requires more computation, but produces more consistent accuracy improvements across models and datasets.