Self-Evaluating LLMs for Multi-Step Tasks: Stepwise Confidence Estimation for Failure Detection
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866917071336505344 |
|---|---|
| author | Mavi, Vaibhav Jaroria, Shubh Sun, Weiqi |
| author_facet | Mavi, Vaibhav Jaroria, Shubh Sun, Weiqi |
| contents | Reliability and failure detection of large language models (LLMs) is critical for their deployment in high-stakes, multi-step reasoning tasks. Prior work explores confidence estimation for self-evaluating LLM-scorer systems, with confidence scorers estimating the likelihood of errors in LLM responses. However, most methods focus on single-step outputs and overlook the challenges of multi-step reasoning. In this work, we extend self-evaluation techniques to multi-step tasks, testing two intuitive approaches: holistic scoring and step-by-step scoring. Using two multi-step benchmark datasets, we show that stepwise evaluation generally outperforms holistic scoring in detecting potential errors, with up to 15% relative increase in AUC-ROC. Our findings demonstrate that self-evaluating LLM systems provide meaningful confidence estimates in complex reasoning, improving their trustworthiness and providing a practical framework for failure detection. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_07364 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Self-Evaluating LLMs for Multi-Step Tasks: Stepwise Confidence Estimation for Failure Detection Mavi, Vaibhav Jaroria, Shubh Sun, Weiqi Machine Learning Artificial Intelligence Computation and Language Reliability and failure detection of large language models (LLMs) is critical for their deployment in high-stakes, multi-step reasoning tasks. Prior work explores confidence estimation for self-evaluating LLM-scorer systems, with confidence scorers estimating the likelihood of errors in LLM responses. However, most methods focus on single-step outputs and overlook the challenges of multi-step reasoning. In this work, we extend self-evaluation techniques to multi-step tasks, testing two intuitive approaches: holistic scoring and step-by-step scoring. Using two multi-step benchmark datasets, we show that stepwise evaluation generally outperforms holistic scoring in detecting potential errors, with up to 15% relative increase in AUC-ROC. Our findings demonstrate that self-evaluating LLM systems provide meaningful confidence estimates in complex reasoning, improving their trustworthiness and providing a practical framework for failure detection. |
| title | Self-Evaluating LLMs for Multi-Step Tasks: Stepwise Confidence Estimation for Failure Detection |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2511.07364 |