- Rating
- 2
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper argues that activation steering should act only at the few high-entropy tokens where a math solution branches, instead of adding one fixed vector at every token. It proposes Hyperbolic Entropy Steering (HEST): a probe trained without correctness labels maps hidden states into a Poincare ball so that a readout orders them by the model's own next-token entropy, and at these "hesitation points" HEST moves the state along the geodesic that lowers the readout and decodes the next token from the edited state.
On three models, HEST changes greedy accuracy on MATH-500 and GSM8K by -0.30 to +1.80 points while steering a median of 0.8-3.0% of the tokens per solution, with gains in five of six settings. CAA added at every token lowers accuracy in all six settings. Per-setting sign tests give p = 0.09-0.64. On the development model, a Euclidean probe lowers accuracy, which the authors take as evidence that the gain depends on the hyperbolic geometry.
Critical strengths
-
Choosing where to steer by entropy is effective, and picking these positions needs no correctness labels. On the development model, edits of the same size give 76.00 (+1.80) on MATH-500 at hesitation points, but applied at every token or at step boundaries they fall below greedy decoding, to 74.00 (-0.20) and 73.80 (-0.40). In all six settings, within the group with 0-10 hesitation points, which holds 71-81% of the problems, HEST turns at most 3% of the correctly solved problems into wrong ones. Unlike the common practice of adding the same vector at every token, this shows that where to intervene matters.
-
The ablations and reporting are thorough. The paper varies where to steer, the direction (probe, fixed, reverse) and the geometry (Euclidean probe), and checks that the edited states realize the step of Proposition 1 to within 10^-4 on average.
Critical weaknesses
-
The gains are small and not statistically significant. Per-setting sign tests give p = 0.09-0.64, and the pooled p = 0.050 treats six settings that share problems as independent. Summed over the four settings other than the development model, whose step length was chosen on the test sets, HEST fixes 49 problems and breaks 40, almost the same number. Each setting uses a single probe, so the variation across probe training runs is unknown.
-
The direction controls do not show that the gain comes from the probe direction. Under the same gate and edit size, the fixed direction, which uses the CAA direction as is, also reaches 76.00 (+1.80) on MATH-500 (13/4 fixed/broken, against 16/7 for HEST), and the reverse direction also gains +0.40 and +0.15 points on MATH-500 and GSM8K. The fixed direction gains less on GSM8K (+0.08 vs. +0.53), but over both benchmarks only the probe direction is right on 39 problems and only the fixed direction on 33. No random-direction control at hesitation points is reported.
-
The only external baseline is CAA, which differs from HEST in edit size and position. CAA uses multiplier 1, and on the development model its vector is about five times the size of a HEST edit and is added at every token. HEST is not compared with HARP [1], which also acts at high-entropy tokens, or with simple decoding alternatives at hesitation points.
[1] R. Storai and S. Hwang. HARP: Hesitation-aware reframing in transformer inference pass. NAACL, 2025.
Questions for authors
-
Statistical robustness. Please report the mean +/- std of the accuracy change over 3-5 probes per setting trained with different seeds. Please also select the step length on held-out training problems rather than the test sets, and evaluate on a larger pool, such as the remaining MATH test problems not included in MATH-500. If the gains hold under these conditions, weakness 1 would be largely addressed.
-
Random-direction control. Please run a random direction (several seeds) and the fixed direction with the same gate, KV rollback and edit size in all six settings. The fixed direction is currently reported only for the development model. Weakness 2 would be addressed if HEST clearly beats both.
-
Mechanism at the steered token. At what fraction of steered positions does the edit change the argmax token, and does the model's actual next-token entropy at that position decrease? How do these quantities differ between the probe, fixed and reverse directions?
10 comments
Response to Reviewer pVhs (2/5): W1 and Q1, part 2
Q1. [...] and evaluate on a larger pool, such as the remaining MATH test problems not included in MATH-500. If the gains hold under these conditions, weakness 1 would be largely addressed.
Answer: We thank the reviewer for this suggestion, which has given the paper its strongest result. We run the development model on the full MATH test set of 5,000 problems, with the probe and the validation-selected step length t = 0.25 unchanged (Section 4.3, Figure 2, Appendix H.2). The revised paper has a new section and a new figure for this experiment.
Overall result and the remaining problems. The results on all 5,000 problems, and on the 4,500 problems that are not in MATH-500, are as follows.
| Problems | Greedy | HEST | Fixed / broken | Exact two-sided sign test |
|---|---|---|---|---|
| Full MATH test set, 5,000 problems | 75.82 | 76.78 | 134 / 86 | p = 0.0015 |
| Remaining problems not in MATH-500, 4,500 problems | 76.00 | 76.98 | 125 / 81 | p = 0.0026 |
The improvement is statistically significant on the full test set, and it is equally large and also significant on the remaining problems that are not included in MATH-500. The gain is therefore not specific to MATH-500.
Breakdown of the test set. Table 10 of the revised paper divides the 5,000 problems by difficulty level, by subject and by the number of hesitation points in the greedy solution, and Figure 2 shows the same numbers. They are as follows.
| Part of the MATH test set | Problems | Greedy | HEST | Fixed / broken |
|---|---|---|---|---|
| Level 1 | 437 | 94.74 | 96.34 | 8 / 1 |
| Level 2 | 894 | 89.04 | 89.82 | 11 / 4 |
| Level 3 | 1131 | 84.88 | 86.03 | 24 / 11 |
| Level 4 | 1214 | 73.89 | 74.63 | 32 / 23 |
| Level 5 | 1324 | 54.68 | 55.59 | 59 / 47 |
| Algebra | 1187 | 91.66 | 92.25 | 15 / 8 |
| Prealgebra | 871 | 85.76 | 86.45 | 18 / 12 |
| Number theory | 540 | 79.26 | 80.37 | 15 / 9 |
| Counting and probability | 474 | 69.83 | 71.52 | 19 / 11 |
| Geometry | 479 | 64.51 | 65.97 | 13 / 6 |
| Precalculus | 546 | 62.45 | 64.10 | 18 / 9 |
| Intermediate algebra | 903 | 60.58 | 61.13 | 36 / 31 |
| 0 to 10 hesitation points | 3640 | 89.73 | 90.25 | 48 / 29 |
| 11 to 24 hesitation points | 910 | 47.14 | 49.45 | 57 / 36 |
| At least 25 hesitation points | 450 | 21.33 | 23.11 | 29 / 21 |
| All problems | 5000 | 75.82 | 76.78 | 134 / 86 |
Accuracy increases in each of these 15 parts, at every difficulty level and in every subject.
Where the gain comes from (lines 400 to 410, Figure 3, Table 11). The last three rows divide the problems by the number of hesitation points in their greedy solution, which is fixed before any intervention. The group with 0 to 10 hesitation points contains 73% of the problems, and greedy decoding already solves 89.73% of them. HEST raises the accuracy of the two groups that hesitate more from 47.14% to 49.45% and from 21.33% to 23.11%. The same pattern holds in the six settings of Table 2: summed over all settings and probes, HEST fixes 81 problems and breaks 33 in the group with at least 25 hesitation points. HEST thus leaves the problems that greedy decoding already solves almost unchanged and improves those on which the model hesitates often.
With the step length selected on held-out problems, three probes per setting and the larger pool, the gains hold under all three conditions that the reviewer names in Q1.
Response to Reviewer pVhs (3/5): W2 and Q2
W2. The direction controls do not show that the gain comes from the probe direction. Under the same gate and edit size, the fixed direction, which uses the CAA direction as is, also reaches 76.00 (+1.80) on MATH-500 (13/4 fixed/broken, against 16/7 for HEST), and the reverse direction also gains +0.40 and +0.15 points on MATH-500 and GSM8K. The fixed direction gains less on GSM8K (+0.08 vs. +0.53), but over both benchmarks only the probe direction is right on 39 problems and only the fixed direction on 33. No random-direction control at hesitation points is reported.
Q2. Random-direction control. Please run a random direction (several seeds) and the fixed direction with the same gate, KV rollback and edit size in all six settings. The fixed direction is currently reported only for the development model. Weakness 2 would be addressed if HEST clearly beats both.
Answer: We thank the reviewer for this request. The submission reported the fixed direction for the development model only and had no random direction, so it could not show that the gain comes from the direction of the probe. We have run the controls exactly as the reviewer specifies (Section 5, Table 3, Appendix J.1).
Setup (lines 1350 to 1361). Every control keeps the layer, the threshold, the gate, the KV rollback and the fallback of HEST and changes only the edit. The fixed direction is the unit CAA direction. A random direction is a vector with independent standard normal entries in the hidden space, normalized to unit length, and we draw three of them with different random seeds. The reverse direction is HEST with the step taken away from the ideal point. In each setting, the fixed and the random directions have the median relative edit size of HEST at its validation-selected step length. We run all controls in all six settings.
Results. Table 3 of the revised paper reports the test accuracy of every variant, and Table 15 in Appendix J.1 gives every control run with its numbers of problems fixed and broken. HEST is run here with the probe that selects the step length, which is the probe that the controls use. The random column is the mean and standard deviation over the three random directions. The accuracies are as follows.
| Model | Dataset | Greedy | HEST | Fixed | Euclidean | Reverse | Random, mean std |
|---|---|---|---|---|---|---|---|
| Qwen2.5-Math-1.5B | MATH-500 | 74.20 | 75.00 | 74.80 | 73.80 | 74.20 | 74.20 0.35 |
| Qwen2.5-Math-1.5B | GSM8K | 84.84 | 85.29 | 84.76 | 84.69 | 84.46 | 84.56 0.63 |
| Llama-3.1-8B | MATH-500 | 50.80 | 52.00 | 49.40 | 50.40 | 50.20 | 50.20 0.92 |
| Llama-3.1-8B | GSM8K | 84.61 | 85.06 | 84.23 | 84.08 | 84.31 | 84.53 0.23 |
| Qwen2.5-Math-7B | MATH-500 | 84.00 | 84.60 | 83.80 | 84.20 | 82.20 | 84.33 0.12 |
| Qwen2.5-Math-7B | GSM8K | 95.38 | 95.68 | 95.53 | 95.38 | 95.30 | 95.58 0.36 |
HEST against the fixed and the random directions. HEST has the highest accuracy in all six settings (lines 475 to 483). It is the only variant whose mean accuracy over the six settings is above that of greedy decoding, and the only one that fixes more problems than it breaks. The fixed direction is below greedy decoding in four of the six settings. The mean accuracy of the random directions over the six settings is 78.90%, which is close to the 78.97% of greedy decoding, so a small edit at hesitation points is close to neutral in a random direction.
The reverse direction no longer gains. The reviewer notes that in the submission the reverse direction also gained on the development model. With the step length selected on held-out problems, the reverse direction does not exceed greedy decoding in any of the six settings, and it fixes 69 problems and breaks 91. Stepping along the direction of the probe improves accuracy, and stepping against it lowers it.
Conclusion. With the same gate, the same KV rollback and the same edit size, neither a fixed direction nor random directions give the gain of HEST. The direction of the probe is the one that gives a consistent gain. We have added this conclusion to Section 5 and to the conclusion of the paper.
Response to Reviewer pVhs (1/5): W1 and Q1, part 1
We sincerely thank the reviewer for the detailed and constructive review, and in particular for stating precisely which experiments would address each concern. We have run all of them, and we believe the paper is considerably stronger as a result. All new results are marked in blue in the revised PDF, and the line, table and appendix numbers below refer to it.
Answer: We fully agree with this concern. In the submission, the step length of the development model was chosen on the test sets, and every setting used a single probe, so the variation across probe training runs was unknown. We have rerun the evaluation exactly as the reviewer suggests.
1. The step length is selected on held-out training problems (lines 287 to 291, Appendix G.2). For every model and dataset, we draw 500 problems from the training split that are disjoint from the 100 problems of the probe and from the test sets. The MATH validation set is stratified by difficulty level to match MATH-500. We run HEST on these problems with each step length in {0.1, 0.25, 0.5, 1} and select the one with the highest accuracy. The thresholds are unchanged and are calibrated without labels, so no test label enters any hyperparameter. Table 1 of the revised paper reports this selection, and its numbers of correctly solved validation problems are as follows.
Within a setting, the four step lengths differ by only 3 to 8 of the 500 problems, so HEST is robust to this choice.
2. Mean and standard deviation over three probes (Table 2, and Table 9 in Appendix H.1). We train three probes per setting with different random seeds and evaluate each of them on the test set at the selected step length. The accuracy of every probe, together with the mean and the standard deviation, is shown below.
HEST improves the mean accuracy of greedy decoding in all six settings, with a standard deviation of 0.12 to 0.42 points across probes, and 16 of the 18 individual runs are at or above greedy decoding (lines 337 to 343).
The evaluation on a larger pool of problems, which is the third request of Q1, is reported in the next comment.