- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
The paper explores prompt sensitivity of LLMs in generating answers. The authors put forward cases where slight changes in prompt formulation, specificity and sampling might drastically change the answer produced. They discuss three separate factors: correctness, formulation dependence and output dispersion to evaluate prompt sensitivity. They provide a new metric that measures how much the correctness of an LLMs answer depends on the prompt formulation. The authors experiment on the AmbigQA dataset showing the contribution of each of the factors on different aspects of prompt sensitivity.
Critical strengths
- The motivation feels decently well portrayed. With LLMs used in many parts of life, prompt sensitivity analysis is a relevant research problem.
- The authors perform a detailed analysis of ten existing metrics.
Critical weaknesses
- The experimentation and generalization of the findings is limited. The authors experiment only on the AmbigQA dataset which mainly contains short factual questions. Hence it can not be confidently established whether the claims hold true for reasoning or summarization tasks as well.
- Experiments have been performed only on the relatively small open-weight models.
- The proposed metric is conceptually similar to the existing metric . The contribution of this newly proposed metric seems to be somewhat incremental.
Questions for authors
- Is it possible to experiment on other datasets alongside AmbigQA, also on other tasks?
- Does the sensitivity values hold true for large weight models as well?
- Could the authors clarify what advantages provides over adapting existing variance-based metrics, particularly given that simpler metrics remain competitive in some of the experiments.
2 comments
Thank you again for your thoughtful review. If you have a chance to look at our revision and response, we’d welcome your thoughts on whether they address your concerns.
Reply to Reviewer Mdaw
We thank you for the constructive review. Our general comment lists what we added, how AmbigQA answers are scored, and the results per model. We address your points in the following:
We added AMBROSIA, a text-to-SQL benchmark from which we use 600 questions, as a second dataset. As on AmbigQA, disambiguation raises accuracy on AMBROSIA (by .23 to .49) and lowers semantic entropy Hsem in all seven models. More varied formulations raise the hierarchical estimate of ρF on AMBROSIA in six of the seven models. They change accuracy by less than one percentage point, with intervals that include zero, and Hsem by less than .03. Therefore, we replicated two results on a different task. Accuracy and output dispersion respond to disambiguation, and ρF responds to more varied formulations. Reasoning and open-ended tasks such as summarization remain untested (Section 6). Our general comment reports how ρF changes under disambiguation on both datasets. On AMBROSIA's ambiguous questions, 29 to 58 % of the variance between formulations is associated with which annotated interpretation the answers take, against 5 to 15 % after disambiguation. We report these results in Section 4.3 and in Appendix D.7.
We added four larger open-weight models and replicated the effect of disambiguation on accuracy and semantic entropy in them. Disambiguation raises accuracy and lowers Hsem in all seven models from 7B to 72B on both datasets (table in our general comment). More varied formulations raise ρF in three of the four larger models on each dataset. On AmbigQA, it falls in Gemma-4-31B (by .03), and on AMBROSIA, a rise is not detected in Qwen2.5-72B. The level of ρF differs between models (mean .10 to .68 across the four larger models on AmbigQA). We report these runs in Section 4.3 and in Appendices D.2 and D.8.
We compared ρF with ρu on the answers of three models and found two differences. We agree that the two metrics are conceptually similar. Both are shares of variance between formulations, and both rank the models in the same order. First, the embedding variance share ρu is computed on answer embeddings, not on correctness. In 25 to 36 % of AmbigQA cells, all answers are correct or all answers are incorrect, although the answers still differ. In these cells, ρu averages .17 to .38. Second, ρu does not subtract decoding noise. When we shuffle the answers of a cell across formulations, no formulation effect remains. ρu then still averages .09 at ten formulations and ten samples, whereas the hierarchical estimate of ρF stays within .01 of zero on AmbigQA. Therefore, ρF complements ρu. ρF measures the share of the variation in correctness that depends on the formulation, net of decoding noise.
We tested the three properties that ρF adds over simpler variance-based metrics, on the AmbigQA answers of the same three models. ρF is computed on correctness, it is close to zero when the formulation has no effect, and it depends less on the number of formulations and samples. At a fixed number of formulations and samples, ρF ranks cells in the same way as the plain between-formulation share η2 (ρu adapted to correctness). When we shuffle outcomes across formulations, η2 averages .47 at two samples per formulation, whereas the method-of-moments estimate of ρF, before truncation at zero, stays within .01 of zero. From two to twenty samples per formulation, the hierarchical estimate of ρF falls by 4 to 20 %, whereas ρu falls by 31 to 68 % on the same cells. From four to fifty formulations, ρF stays within .04, whereas the spread rises by .16 to .20. Therefore, we recommend ρF to measure how much the correctness of an answer depends on the formulation. To locate the questions where a change of formulation gains most, we recommend the variance between formulations σB2 or the sampled spread. Section 4.3 states this comparison, and Table 10 in Appendix D.4 compares the three metrics criterion by criterion.
Please let us know if anything remains unclear.