- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper investigates how to make LLMs say "I don't know." Existing methods either train the model to abstain directly (e.g., TruthRL) or rely on an external confidence score to filter out uncertain answers after generation. Noting the lack of a fair comparison between the two, the authors compare their error rates at a fixed answer rate. They also break down the overall error into the model's base accuracy and its ability to selectively withhold answers. They find that if a good confidence estimator is used, both approaches are equally effective at choosing what to withhold; the main benefit of abstention training actually comes from improved base accuracy. Following this finding, the authors propose TAPS, which trains the model purely for accuracy and leaves the abstention decision to an existing confidence method. This approach performs on par with abstention training and keeps the answer rate flexible.
Critical strengths
- The motivation is clear and the definitions are careful. The paper sets up a fair way to compare the two kinds of abstention: it fixes the answer rate and then compares error rates. It also splits the error rate into two parts, how well the model answers and how well it chooses what to withhold. I found this a useful way to think about abstention, and I think other papers in this area could use the same protocol.
Critical weaknesses
- The gain over TruthRL is small, and it doesn't hold up across models. TAPS with probing beats TruthRL by 1.49 pp on average, but almost all of that comes from Qwen3-8B (+5.06 pp). On Qwen2.5-7B, TruthRL is actually better on average (−0.70 pp), and on Llama the two are about the same (+0.12 pp). There are no multiple seeds or confidence intervals, so I can't tell how much of this is noise.
- TAPS itself is not new as a method, and I see it mainly as supporting evidence for the paper's analysis. But for an analysis paper, the set of compared methods is quite narrow: only three representative methods are included. Several newer methods have built on TruthRL since then, such as TIAR (Pan et al., 2026), and KARL (Gao et al., 2026). I would like to see these methods added to the comparison and the analysis, which would make the conclusions more convincing.
Questions for authors
- SE needs 10 extra samples per question, plus entailment clustering. Probing needs labeled data for every model and dataset, and access to hidden states. TruthRL needs only one generation. Could you report the actual inference and training cost?
5 comments
Rebuttal by Authors (Part 2 of 2)
Q1: Training and inference cost
TAPS (Probing) does not add much cost over TruthRL to train and run. But to change coverage, TAPS only moves a threshold, while TruthRL must retrain for another 64 hours.
We measure costs on Qwen3-8B with 4 A100 80GB GPUs for training and inference (a separate set of 4 GPUs serves the judge), using the hyperparameters in Appendix C.4. The reported times are wall-clock durations recorded during our experiments and should be read as approximate.
(1) Training: one TAPS run costs about as much as one TruthRL run, but TruthRL must retrain for every target coverage. Both methods are trained for 8 epochs (624 steps for NQ and TriviaQA, and 312 steps for PopQA). The probe is fitted once on the trained model's outputs, adding a one-time 0.44 hours on average, including answer generation, labeling and fitting.
| Training cost (hours) | NQ | PopQA | TriviaQA | Avg. |
|---|---|---|---|---|
| TruthRL | 81.47 | 40.39 | 71.41 | 64.42 |
| TAPS (training) | 82.68 | 39.87 | 70.89 | 64.48 |
| TAPS (probe fitting) | 0.50 | 0.29 | 0.52 | 0.44 |
(2) Inference: TAPS (Probing) is as fast as TruthRL (19.4 vs. 19.6 min), while TAPS(Semantic Entropy) costs hours due to 10 additional samples per question and entailment clustering.
| Inference time on the test set (min) | NQ | PopQA | TriviaQA | Avg. |
|---|---|---|---|---|
| TruthRL | 9.5 | 20.3 | 28.9 | 19.6 |
| TAPS (Vanilla) | 9.1 | 23.0 | 29.5 | 20.5 |
| TAPS (Probing) | 9.2 | 21.6 | 27.4 | 19.4 |
| TAPS (Semantic Entropy) | 386.9 | 1043.5 | 1255.3 | 895.2 |
Overall, TAPS (Probing) adds little cost over TruthRL in a single run, and saves a full retraining every time the target coverage changes.
These results have been added to Appendix K.
Thank you for your constructive suggestions. With seed-robust results, the comparison with TIAR, and the measured costs, we would be grateful if you would consider raising your score. We are happy to provide any additional details.
Rebuttal by Authors (Additional Results)
Additional experiments further confirm that our conclusions are robust across seeds on all three models.
In our earlier response, we reported multi-seed results on Qwen3-8B. We have now completed the same multi-seed experiment on the other two models: Qwen2.5-7B-Instruct and Llama-3.1-8B-Instruct. In every setting, the direction of the comparison matches the single-run result in our paper (Table 1), suggesting that TAPS consistently achieves comparable or better abstention performance than abstention training, while keeping coverage adjustable. Below are the detailed results.
For each of three seeds, we retrain TruthRL and TAPS, and match TAPS to TruthRL's native coverage to compare risk. We report mean ± std over seeds. Δ = TruthRL risk − TAPS risk (positive favors TAPS).
| Model | Dataset | Coverage | TruthRL | TAPS (Probing) | Δ | 95% Confidence Interval of Δ |
|---|---|---|---|---|---|---|
| Qwen2.5-7B-Instruct | NQ | 49.34 ± 1.24 | 40.65 ± 0.96 | 37.93 ± 0.68 | 2.72 | [0.55, 4.90] |
| PopQA | 20.33 ± 0.57 | 29.41 ± 0.34 | 29.01 ± 0.27 | 0.40 | [−0.28, 1.07] | |
| TriviaQA | 72.98 ± 0.37 | 16.94 ± 0.63 | 17.13 ± 0.60 | −0.19 | [−2.99, 2.60] | |
| Qwen3-8B | NQ | 48.36 ± 1.91 | 36.67 ± 0.52 | 33.54 ± 0.53 | 3.13 | [0.94, 5.32] |
| PopQA | 20.50 ± 0.50 | 32.55 ± 0.97 | 26.16 ± 1.04 | 6.39 | [6.06, 6.71] | |
| TriviaQA | 78.24 ± 0.40 | 15.93 ± 0.59 | 14.67 ± 0.40 | 1.26 | [0.26, 2.27] | |
| Llama-3.1-8B-Instruct | NQ | 76.78 ± 0.73 | 33.23 ± 0.39 | 34.51 ± 0.59 | −1.28 | [−1.87, −0.69] |
| PopQA | 43.67 ± 0.98 | 27.86 ± 0.53 | 29.31 ± 1.50 | −1.44 | [−4.60, 1.72] | |
| TriviaQA | 90.46 ± 0.79 | 11.23 ± 0.46 | 10.96 ± 0.26 | 0.27 | [−0.25, 0.80] | |
| Avg. | 27.16 | 25.91 | 1.25 | [0.93, 1.57] |
We would greatly appreciate it if you could let us know whether our responses have addressed your concerns. We are happy to discuss anything further if helpful.
Rebuttal by Authors (Part 1 of 2)
Dear reviewer pXij,
Thank you for your detailed and constructive review. We are grateful that you acknowledge the clear motivation and careful definitions, the fair comparison protocol, and the useful decomposition that other papers in this area could adopt.
W1: Robustness across seeds
(1) Since TAPS retains the coverage control that abstention training lacks, a large margin over these methods is not our aim. Performing on par with the strongest of them, TruthRL, is already enough for TAPS to be the stronger deployment choice, and it does slightly better. Our evaluation is not limited to a single setting: it spans 9 independently trained model–dataset settings and three abstention rewards for TruthRL, so the conclusion does not hinge on any one run.
(2) To check that this is not seed noise, we repeat the full Qwen3-8B pipeline with 3 seeds and find TAPS's performance against TruthRL statistically significant on all three datasets (p≤0.033, paired t-test). Each seed trains its own TruthRL (abstention reward set to 0) and TAPS checkpoints and refits the probe. TAPS is matched to the coverage of the same-seed TruthRL checkpoint.
W2: Comparison with recent abstention training methods
Our study covers SFT-based (R-Tuning), reward-based (TruthRL under three abstention rewards), and advantage-based (UCPO) abstention training. Following your suggestion, we further consider the two recent methods:
(1) TAPS achieves clearly lower risk than TIAR, and TIAR shows the same pattern as TruthRL in terms of decomposition. For a controlled comparison, we implement TIAR's published objective in our shared verl-based framework with its default method-specific hyperparameters, and train and evaluate it on Qwen3-8B under exactly the same setup as all other methods on all three datasets.
a. Comparison with TAPS. At TIAR's native coverage, TAPS (Probing) achieves 22.32% risk on average, compared with 28.87% for TIAR.
b. Decomposition. To see where TIAR's performance comes from, we repeat the decomposition analysis in Section 4, comparing TIAR with selective answering using semantic entropy (S.E.) on the starting model at TIAR's native coverage (%):
TIAR nearly matches S.E. in distinguishing gain, so its standing depends on answering gain. On NQ and TriviaQA, its answering gain is positive and its risk is lower. On PopQA, its answering gain is close to zero and cannot offset even a small distinguishing gap, so its risk is higher.
(2) KARL has not released code or checkpoints, so we could not include it within the rebuttal period. As an RL-reward method like TruthRL and UCPO, it can be analyzed with the same protocol once available.