The Self-Improvement Paradox: Can Language Models Bootstrap Reasoning Capabilities without External Scaffolding?
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866915160734564352 |
|---|---|
| author | Sun, Yutao Chen, Mingshuai Zhao, Tiancheng Xu, Ruochen Zhang, Zilun Yin, Jianwei |
| author_facet | Sun, Yutao Chen, Mingshuai Zhao, Tiancheng Xu, Ruochen Zhang, Zilun Yin, Jianwei |
| contents | Self-improving large language models (LLMs) -- i.e., to improve the performance of an LLM by fine-tuning it with synthetic data generated by itself -- is a promising way to advance the capabilities of LLMs while avoiding extensive supervision. Existing approaches to self-improvement often rely on external supervision signals in the form of seed data and/or assistance from third-party models. This paper presents Crescent -- a simple yet effective framework for generating high-quality synthetic question-answer data in a fully autonomous manner. Crescent first elicits the LLM to generate raw questions via a bait prompt, then diversifies these questions leveraging a rejection sampling-based self-deduplication, and finally feeds the questions to the LLM and collects the corresponding answers by means of majority voting. We show that Crescent sheds light on the potential of true self-improvement with zero external supervision signals for math reasoning; in particular, Crescent-generated question-answer pairs suffice to (i) improve the reasoning capabilities of an LLM while preserving its general performance (especially in the 0-shot setting); and (ii) distil LLM knowledge to weaker models more effectively than existing methods based on seed-dataset augmentation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_13441 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | The Self-Improvement Paradox: Can Language Models Bootstrap Reasoning Capabilities without External Scaffolding? Sun, Yutao Chen, Mingshuai Zhao, Tiancheng Xu, Ruochen Zhang, Zilun Yin, Jianwei Computation and Language Artificial Intelligence Self-improving large language models (LLMs) -- i.e., to improve the performance of an LLM by fine-tuning it with synthetic data generated by itself -- is a promising way to advance the capabilities of LLMs while avoiding extensive supervision. Existing approaches to self-improvement often rely on external supervision signals in the form of seed data and/or assistance from third-party models. This paper presents Crescent -- a simple yet effective framework for generating high-quality synthetic question-answer data in a fully autonomous manner. Crescent first elicits the LLM to generate raw questions via a bait prompt, then diversifies these questions leveraging a rejection sampling-based self-deduplication, and finally feeds the questions to the LLM and collects the corresponding answers by means of majority voting. We show that Crescent sheds light on the potential of true self-improvement with zero external supervision signals for math reasoning; in particular, Crescent-generated question-answer pairs suffice to (i) improve the reasoning capabilities of an LLM while preserving its general performance (especially in the 0-shot setting); and (ii) distil LLM knowledge to weaker models more effectively than existing methods based on seed-dataset augmentation. |
| title | The Self-Improvement Paradox: Can Language Models Bootstrap Reasoning Capabilities without External Scaffolding? |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2502.13441 |