Test-time Scaling Techniques in Theoretical Physics -- A Comparison of Methods on the TPBench Dataset
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866915359291867136 |
|---|---|
| author | Gao, Zhiqi Li, Tianyi Kvasiuk, Yurii Tadepalli, Sai Chaitanya Rudolph, Maja Chung, Daniel J. H. Sala, Frederic Münchmeyer, Moritz |
| author_facet | Gao, Zhiqi Li, Tianyi Kvasiuk, Yurii Tadepalli, Sai Chaitanya Rudolph, Maja Chung, Daniel J. H. Sala, Frederic Münchmeyer, Moritz |
| contents | Large language models (LLMs) have shown strong capabilities in complex reasoning, and test-time scaling techniques can enhance their performance with comparably low cost. Many of these methods have been developed and evaluated on mathematical reasoning benchmarks such as AIME. This paper investigates whether the lessons learned from these benchmarks generalize to the domain of advanced theoretical physics. We evaluate a range of common test-time scaling methods on the TPBench physics dataset and compare their effectiveness with results on AIME. To better leverage the structure of physics problems, we develop a novel, symbolic weak-verifier framework to improve parallel scaling results. Our empirical results demonstrate that this method significantly outperforms existing test-time scaling approaches on TPBench. We also evaluate our method on AIME, confirming its effectiveness in solving advanced mathematical problems. Our findings highlight the power of step-wise symbolic verification for tackling complex scientific problems. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_20729 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Test-time Scaling Techniques in Theoretical Physics -- A Comparison of Methods on the TPBench Dataset Gao, Zhiqi Li, Tianyi Kvasiuk, Yurii Tadepalli, Sai Chaitanya Rudolph, Maja Chung, Daniel J. H. Sala, Frederic Münchmeyer, Moritz Machine Learning Cosmology and Nongalactic Astrophysics Artificial Intelligence High Energy Physics - Phenomenology High Energy Physics - Theory Large language models (LLMs) have shown strong capabilities in complex reasoning, and test-time scaling techniques can enhance their performance with comparably low cost. Many of these methods have been developed and evaluated on mathematical reasoning benchmarks such as AIME. This paper investigates whether the lessons learned from these benchmarks generalize to the domain of advanced theoretical physics. We evaluate a range of common test-time scaling methods on the TPBench physics dataset and compare their effectiveness with results on AIME. To better leverage the structure of physics problems, we develop a novel, symbolic weak-verifier framework to improve parallel scaling results. Our empirical results demonstrate that this method significantly outperforms existing test-time scaling approaches on TPBench. We also evaluate our method on AIME, confirming its effectiveness in solving advanced mathematical problems. Our findings highlight the power of step-wise symbolic verification for tackling complex scientific problems. |
| title | Test-time Scaling Techniques in Theoretical Physics -- A Comparison of Methods on the TPBench Dataset |
| topic | Machine Learning Cosmology and Nongalactic Astrophysics Artificial Intelligence High Energy Physics - Phenomenology High Energy Physics - Theory |
| url | https://arxiv.org/abs/2506.20729 |