Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answering
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909694334861312 |
|---|---|
| author | Jurayj, William Cheng, Jeffrey Van Durme, Benjamin |
| author_facet | Jurayj, William Cheng, Jeffrey Van Durme, Benjamin |
| contents | Scaling the test-time compute of large language models has demonstrated impressive performance on reasoning benchmarks. However, existing evaluations of test-time scaling make the strong assumption that a reasoning system should always give an answer to any question provided. This overlooks concerns about whether a model is confident in its answer, and whether it is appropriate to always provide a response. To address these concerns, we extract confidence scores during reasoning for thresholding model responses. We find that increasing compute budget at inference time not only helps models answer more questions correctly, but also increases confidence in correct responses. We then extend the current paradigm of zero-risk responses during evaluation by considering settings with non-zero levels of response risk, and suggest a recipe for reporting evaluations under these settings. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_13962 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answering Jurayj, William Cheng, Jeffrey Van Durme, Benjamin Computation and Language Scaling the test-time compute of large language models has demonstrated impressive performance on reasoning benchmarks. However, existing evaluations of test-time scaling make the strong assumption that a reasoning system should always give an answer to any question provided. This overlooks concerns about whether a model is confident in its answer, and whether it is appropriate to always provide a response. To address these concerns, we extract confidence scores during reasoning for thresholding model responses. We find that increasing compute budget at inference time not only helps models answer more questions correctly, but also increases confidence in correct responses. We then extend the current paradigm of zero-risk responses during evaluation by considering settings with non-zero levels of response risk, and suggest a recipe for reporting evaluations under these settings. |
| title | Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answering |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2502.13962 |