- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes a two-stage framework for calibrating verbalized confidence in large language models using multiple sources of evidence. In the first stage, the model's parametric knowledge and retrieved documents are converted into evidence representations. They are aggregated through an attention-based evidence fusion (AEF) module to parameterize a Dirichlet distribution over answer options. In the second stage, the trained AEF module generates soft confidence targets using only the model's logits, which are conditioned by the question, without requiring document retrieval as in the first stage. These targets are used to fine-tune the LLM through a loss function containing the Brier score, KL divergence, and answer prediction losses.
The method is evaluated on Llama-3.1-8B-Instruct and Qwen3-8B-Instruct, with training on MedQA and OOD evaluation on PubMedQA, TriviaQA, StrategyQA, and TruthfulQA. The reported results show improved average calibration and confidence discrimination compared with existing baselines, particularly on the in-distribution dataset. The paper also provides ablation studies of the training objectives, evidence fusion module, and several hyperparameters.
Critical strengths
S1. Transferring retrieval-informed confidence estimates into verbalized confidence.
The two-stage design provides a plausible framework for using external evidence during confidence estimator training while avoiding additional per-document LLM passes when producing confidence targets during Stage 2. In particular, masked-retrieval training connects the multi-source estimator to the setting without retrieval that supplies soft targets. This is meaningful because it attempts to retain the benefits of retrieval-informed supervision without requiring the full multi-source estimation procedure whenever confidence is expressed. The practical value of this separation is a substantive reason to consider the approach.
S2. Substantial in-distribution performance improvement.
On MedQA, the reported ECE decreases, and AUC improves substantially for Llama and Qwen relative to ConfTuner. Improvements in both calibration and discrimination would be practically meaningful, with the reliability diagram in Figure 2 providing qualitative support for better alignment between confidence and empirical accuracy. These in-distribution results constitute the strongest empirical evidence in the paper, subject to resolution of the evaluation-protocol concerns below.
Critical weaknesses
W1. Insufficient justification for the epistemic uncertainty interpretation.
The paper interprets as epistemic uncertainty, but the link between logit-derived evidence strength and the LLM's knowledge remains insufficiently validated. For the source-level mapping , a common logit shift preserves normalized answer-option probabilities but changes the corresponding evidence strength and uncertainty. The epistemic meaning of this sensitivity needs justification. Moreover, the claimed uncertainty responses to source agreement and conflict are not directly validated by controlled experiments. Proposition 2 establishes variance reduction for fixed teacher, per-example optimization, rather than a calibration guarantee for the fine-tuned LLM. These gaps may weaken the paper's central theoretical motivation.
W2. Insufficient isolation of the retrieval-based evidence fusion contribution.
During Stage 2, all retrieval branches are masked, and the confidence targets depend only on no-retrieval logits. Although Appendix E.2 includes , it does not report the final verbalized confidence calibration results after Stage 2 under this setting, with all other training conditions kept unchanged. Thus, the contribution of retrieval-based evidence learning cannot be separated from general confidence distillation and KL regularization. Furthermore, the retrieval and no-retrieval evaluations in the supplementary implementation use potentially different predicted answers (204/1,268 cases in the supplied Llama test logits), making it difficult to isolate improvements in confidence estimation itself.
W3. Concerns about the evaluation protocol and reliability of reported calibration metrics.
The supplementary code uses different confidence definitions across evaluations. For MedQA, confidence is computed as the expectation of the confidence-token distribution, , whereas OOD evaluation uses the generated confidence value, , where is the generated confidence token. It is unclear which definitions were used for the reported results in Table 1 and whether the improvements hold under a consistent evaluation procedure. The PubMedQA prompt restricts answers to yes/no despite the supplied dataset containing "maybe" labels. These evaluation issues require clarification to establish the reliability of the reported results. Answer accuracy after fine-tuning is also not reported, making it difficult to assess whether calibration improvements preserve answer quality.
Questions for authors
Q1. Epistemic uncertainty interpretation (Sections 3- 4, Appendix C).
How can you justify the interpretation of as epistemic uncertainty despite its sensitivity to common logit shifts? Can the authors provide controlled evidence that the proposed uncertainty measure appropriately responds to supporting, conflicting, or irrelevant retrieved documents?
Q2. Retrieval contribution (Sections 4.2–4.3, Appendix E).
How does final verbalized confidence calibration change when AEF is trained without retrieval, while keeping the architecture and Stage 2 training identical? This may establish whether retrieval-based evidence learning provides benefits beyond general confidence distillation.
Q3. Retrieval versus no-retrieval comparison (Table 3).
Does Table 3 evaluate different predicted answers and correctness labels across retrieval conditions, as suggested by the supplementary implementation? If so, could the authors report results with the predicted answers and correctness labels held fixed to isolate the contribution of retrieval to confidence estimation?
Q4. Confidence extraction and metric computation (Table 1, supplementary code).
The supplementary code appears to use for MedQA but the generated confidence value for OOD datasets. Could the authors clarify which confidence definitions were used for Table 1 and whether the reported improvements hold under a consistent evaluation procedure?
Q5. PubMedQA evaluation protocol (Appendix G.2).
How were "maybe"-labeled PubMedQA examples handled given the yes/no-only evaluation prompt? Were these examples excluded, or was a different prompt used for the reported results? If neither, can the authors provide corrected results?
Q6. Answer prediction accuracy (Section 5, Tables 1–2).
Could the authors report answer accuracy before and after fine-tuning, including the loss ablations? This would clarify whether the calibration improvements preserve answer quality.
7 comments
The details and data for this comment
We have reorganized the detailed data in the response for the reviewers’ direct review.
- Regarding question 1, we agree that a common shift of the answer-option logits preserves their softmax probabilities but can change the evidence obtained through the softplus mapping. Consequently, our formulation does not guarantee shift invariance, and interpreting as a direct measure of the LLM’s epistemic uncertainty was too strong. In our implementation, AEF operates on the native logits of a fixed backbone under a specified prompting and extraction procedure. Common shifts are not explicitly introduced during normal inference. To examine responses to different types of evidence, we constructed a controlled synthetic-evidence experiment. Given each question, its answer options, and the ground-truth answer, we prompted DeepSeek to generate two supporting passages intended to substantiate the correct answer and two misleading passages intended to favor an incorrect alternative. The latter served as a conflicting-evidence condition. We additionally constructed a mismatched condition by assigning passages from other questions, providing a control for irrelevant evidence. Because the synthetic passages were generated with access to the ground-truth answer, this experiment evaluates controlled evidence interventions rather than the quality of a realistic retrieval pipeline. For this supplementary experiment, we trained AEF using supporting passages from the training split. The evaluated answer for each question was fixed to the LLM’s prediction made without retrieval. Predictions and correctness labels therefore remained identical across conditions, with 704 correct and 514 incorrect predictions. Confidence was computed as the Dirichlet mean assigned to that fixed answer.
Table 5. AEF Uncertainty and Confidence under Controlled Evidence Conditions.
| Metric | Without Retrieval | Supporting | Conflicting | Mismatch |
|---|---|---|---|---|
| Average Uncertainty (u) | 0.2781 | 0.2342 | 0.2363 | 0.2390 |
| Confidence of Correct Answers | 0.5966 | 0.6273 | 0.5734 | 0.5612 |
| Confidence of Incorrect Answers | 0.4510 | 0.4101 | 0.4394 | 0.4224 |
Supporting evidence increased mean confidence on correct predictions from 0.5966 to 0.6273 and decreased it on incorrect predictions from 0.4510 to 0.4101. In contrast, misleading and mismatched evidence reduced confidence in both groups. These results demonstrate confidence adjustments consistent with prediction correctness under this supporting-evidence training condition. The interpretation of requires a distinction between evidence strength and evidence quality. Within our Dirichlet parameterization, reflects the total concentration assigned across all answer options, rather than whether the evidence supports the ground-truth answer. Evidence assigned to an incorrect alternative can also increase total concentration and therefore reduce . Accordingly, a lower does not necessarily indicate more reliable evidence or a more accurate prediction. So, all three document conditions reduced mean relative to the no-retrieval baseline.
The details and data for this comment
- Regarding question 2, we added an ablation in which AEF is trained without retrieval, while keeping its architecture and the Stage 2 training procedure unchanged. The no-retrieval AEF is then used as the confidence teacher in Stage 2. The table reports final verbalized confidence calibration and answer accuracy, with averages computed across the five datasets. The results show that incorporating retrieval during AEF training improves the average confidence estimation performance of both models. For Llama-8B, average ECE decreases from 0.1552 to 0.0882, Brier score decreases from 0.2479 to 0.2028, and AUC increases from 0.6027 to 0.6990. Qwen-8B shows smaller but consistent improvements in these average metrics, although its average answer accuracy decreases slightly from 0.6613 to 0.6560.
Table 6. Effect of Retrieval during AEF Training on Final Verbalized Confidence Calibration and Answer Accuracy.
| Metric | Model | Method | Med (ID) | Pub (OOD) | Tri (OOD) | Str (OOD) | Tru (OOD) | Average |
|---|---|---|---|---|---|---|---|---|
| ECE ↓ | Llama 8B | w/o RE | 0.2313 | 0.0487 | 0.0875 | 0.1172 | 0.2914 | 0.1552 |
| ECE ↓ | Llama 8B | Ours | 0.0518 | 0.1008 | 0.0575 | 0.0831 | 0.1478 | 0.0882 |
| ECE ↓ | Qwen 8B | w/o RE | 0.1584 | 0.0487 | 0.2436 | 0.1265 | 0.1116 | 0.1378 |
| ECE ↓ | Qwen 8B | Ours | 0.1275 | 0.0338 | 0.2417 | 0.1432 | 0.0662 | 0.1225 |
| AUC ↑ | Llama 8B | w/o RE | 0.5596 | 0.5885 | 0.6457 | 0.6280 | 0.5915 | 0.6027 |
| AUC ↑ | Llama 8B | Ours | 0.7118 | 0.7234 | 0.6968 | 0.6535 | 0.7093 | 0.6990 |
| AUC ↑ | Qwen 8B | w/o RE | 0.6595 | 0.7513 | 0.7213 | 0.5932 | 0.5587 | 0.6568 |
| AUC ↑ | Qwen 8B | Ours | 0.6787 | 0.7651 | 0.7239 | 0.5936 | 0.6403 | 0.6803 |
| Brier ↓ | Llama 8B | w/o RE | 0.2930 | 0.1802 | 0.2054 | 0.2358 | 0.3251 | 0.2479 |
| Brier ↓ | Llama 8B | Ours | 0.2152 | 0.1700 | 0.1811 | 0.2080 | 0.2397 | 0.2028 |
| Brier ↓ | Qwen 8B | w/o RE | 0.2463 | 0.1646 | 0.2668 | 0.2416 | 0.2014 | 0.2241 |
| Brier ↓ | Qwen 8B | Ours | 0.2313 | 0.1583 | 0.2628 | 0.2493 | 0.1853 | 0.2174 |
| ACC ↑ | Llama 8B | w/o RE | 0.5672 | 0.7510 | 0.6703 | 0.6783 | 0.4774 | 0.6288 |
| ACC ↑ | Llama 8B | Ours | 0.5706 | 0.7487 | 0.7197 | 0.6861 | 0.4851 | 0.6420 |
| ACC ↑ | Qwen 8B | w/o RE | 0.6101 | 0.7270 | 0.5614 | 0.6516 | 0.7564 | 0.6613 |
| ACC ↑ | Qwen 8B | Ours | 0.6104 | 0.7210 | 0.5557 | 0.6536 | 0.7393 | 0.6560 |
- Regarding question 3, specifically, we fixed each evaluated answer to the LLM’s prediction made without retrieval and kept its correctness label unchanged. We compared AEF trained with mixed retrieval exposure () against AEF trained without retrieval (), evaluating each with and without retrieval at test time. For AEF trained with mixed retrieval exposure, adding retrieval at test time leaves the Llama 8B metrics unchanged at the reported precision. For Qwen 8B, ECE changes from 0.0755 to 0.0778, AUC from 0.7317 to 0.7331, and Brier remains 0.2138. In contrast, AEF trained without retrieval shows deterioration in all three metrics when retrieval is introduced at test time. For Qwen 8B, ECE increases from 0.0747 to 0.0973, and AUC decreases from 0.7405 to 0.7261. This comparison suggests that retrieval exposure during AEF training improves the stability of aggregate confidence-estimation metrics across the evaluated test conditions. We include this controlled comparison and revise Appendix E.1 accordingly. In particular, we remove the stronger attribution of the original AUC gains to improved confidence estimation through retrieval and replace it with the more limited interpretation supported by the fixed-answer results.
Table 7. AEF Confidence Estimation with Fixed Predicted Answers across Training and Test Retrieval Conditions.
| Training Condition | Test Condition | Llama 8B ECE ↓ | Llama 8B AUC ↑ | Llama 8B Brier ↓ | Qwen 8B ECE ↓ | Qwen 8B AUC ↑ | Qwen 8B Brier ↓ |
|---|---|---|---|---|---|---|---|
| With Retrieval | Without Retrieval | 0.0473 | 0.7381 | 0.2032 | 0.0755 | 0.7317 | 0.2138 |
| With Retrieval | With Retrieval | 0.0473 | 0.7381 | 0.2032 | 0.0778 | 0.7331 | 0.2138 |
| Without Retrieval | Without Retrieval | 0.0441 | 0.7386 | 0.2032 | 0.0747 | 0.7405 | 0.2116 |
| Without Retrieval | With Retrieval | 0.0454 | 0.7271 | 0.2075 | 0.0973 | 0.7261 | 0.2191 |
Response to Reviewer Comment Point by Point
Thanks for your valuable and careful comment. Below, we answer your questions in detail and outline the corresponding revisions we will make in the final manuscript.
Regarding question 1, we agree that our formulation does not guarantee invariance to common logit shifts and therefore does not fully justify interpreting u=K/S as a direct measure of the LLM’s epistemic uncertainty. We have revised this interpretation to “evidence-based uncertainty” within the adopted Dirichlet parameterization. In our controlled experiment, mean u is highest without retrieval and lower under all three document conditions, with supporting evidence yielding the lowest mean value. However, these results do not establish that u reliably distinguishes evidence quality. We also explicitly acknowledge its sensitivity to common logit shifts as a limitation. The relevant content is shown on page 2-4, 10, 14 and 23 of the revised PDF, in the introduction, the related works, the preliminary, the section 4 , the conclusion, the Appendix B.2 and the Appendix K.
Regarding question 2, training AEF without retrieval worsens average final verbalized confidence calibration for both backbones: average ECE and Brier score increase, while average AUC decreases. With the AEF architecture and Stage 2 training procedure held unchanged, these results support an additional contribution from retrieval-assisted AEF training beyond the confidence-distillation and regularization objectives. The relevant content is shown on page 9 of the revised PDF, in the Table 2 and the paragraph above Table 2.
Regarding question 3, thank you for pointing out this confounding factor. Indeed, the original Table 3 allowed predicted answers and correctness labels to differ across retrieval conditions. We have added the requested comparison with predicted answers and correctness labels held fixed. Under this controlled setting, test-time retrieval provides no clear additional calibration benefit, while AEF trained with retrieval exhibits more stable calibration metrics across test conditions than AEF trained without retrieval. The relevant content is shown on page 16-17 of the revised PDF, in the Appendix E.1 and Table 4.
Regarding question 4 and 5, we thank the reviewer for identifying these evaluation inconsistencies. For Question 4, the original evaluation used the expectation of the confidence-output distribution for MedQA but the generated confidence value for the OOD datasets. We have corrected this inconsistency by using generated confidence values throughout. Under a consistent evaluation procedure, our method retains the lowest average ECE and Brier score among the four main comparison methods for both backbones. For Question 5, “maybe”-labeled examples were retained in the original PubMedQA evaluation, although the prompt incorrectly allowed only “yes” and “no.” We have corrected the prompt to permit all three answers and reevaluated PubMedQA without excluding “maybe”-labeled examples. Although our method outperforms the compared methods on only some metrics on PubMedQA, it achieves the best average ECE, Brier score, and answer accuracy across all five datasets for both backbones. These results indicate that the overall benefits remain supported under the corrected evaluation procedure. Specifically, for methods that generate verbalized confidence, we extract the integer c∈{0,…,9} generated by the LLM and convert it to a confidence score in [0,1] using c/9. The revised corrected results, and accompanying analysis appear on pages 7–9 and 22 of the revised manuscript.
Regarding question 6, our method improves average answer accuracy after QLoRA fine-tuning, from 0.6042 to 0.6420 for Llama 8B and from 0.6089 to 0.6560 for Qwen 8B. Removing the language modeling loss produces the largest decrease in average accuracy among the loss ablations for both backbones. The detailed accuracy results are reported in Appendix G, on pages 19–20 of the revised manuscript.