arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2510.02453cs.LGcs.AIcs.CL
收藏

如何训练你的导师:通过导师模型引导黑盒大语言模型

How to Train Your Advisor: Steering Black-Box LLMs with Advisor Models

University of California, Berkeley(加州大学伯克利分校);Bespoke Labs(Bespoke实验室)

Parth Asawa, Alan Zhu, Abigail O'Neill, Matei Zaharia, Alexandros G. Dimakis, Joseph E. Gonzalez

AI总结:

本文提出Advisor Models,通过训练小型开放权重模型生成动态个性化建议,提升黑盒前沿模型性能,实验显示在多个任务中效果显著,且具有良好的迁移性和鲁棒性。

摘要:

前沿语言模型作为黑盒服务部署,其权重无法修改,定制仅限于提示。我们引入Advisor Models,一种方法通过训练小型开放权重模型生成动态、实例特定的自然语言建议,以提升黑盒前沿模型的能力。Advisor Models将GPT-5.2在RuleArena(税务)任务上的性能提升27.4%,减少Gemini 3 Pro在SWE代理任务中的步骤24.6%,并在个性化GPT-5到用户偏好方面优于静态提示优化器(85-100% vs. 40-60%)。我们还发现顾问具有可迁移性:用低成本学生模型训练的顾问仍能将改进转移到前沿模型。此外,Advisor Models具有鲁棒性:在其他基准测试中未观察到降级,除了训练管道所训练的基准测试。我们的方法展示了如何以实用且经济有效的方式对黑盒前沿模型进行参数优化。

英文摘要:

Frontier language models are deployed as black-box services, where model weights cannot be modified and customization is limited to prompting. We introduce Advisor Models, a method to train small open-weight models to generate dynamic, per-instance natural language advice that improves the capabilities of black-box frontier models. Advisor Models improve GPT-5.2's performance on RuleArena (Taxes) by 27.4%, reduce Gemini 3 Pro's steps taken in SWE agent tasks by 24.6%, and outperform static prompt optimizers in personalizing GPT-5 to user preferences (85-100% vs. 40-60%). We also find that advisors are transferable: an advisor trained with a low-cost student model still transfers improvements to a frontier model. Moreover, Advisor Models are robust: we observe no degradation on other benchmarks than the pipeline is trained on. Our method shows how to perform parametric optimization for black-box frontier models in a practical and cost-effective way.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑