通过基础模型自我探索实现自动化能力发现
Automated Capability Discovery via Foundation Model Self-Exploration
AI总结：
本文提出 ACD 框架，让基础模型作为“科学家”自动生成并聚类开放式任务，以系统发现多种基础模型的能力与失败模式，并通过人类调查验证自动评分的一致性。
摘要：
基础模型已成为通用助手，通过在网络规模数据上训练，在众多领域展现出多样化能力。对于任何新模型，即便要精确刻画其全部能力与潜在风险中的一小部分，仍然颇具挑战。现有评估方法通常需要大量人力，而且为能力更强的模型设计越来越困难的挑战，所需投入也在不断增加。我们提出 Automated Capability Discovery（ACD，自动化能力发现）框架，该框架指定一个基础模型充当科学家，系统性地提出开放式任务，以探测受试模型（可能是其自身）的能力。通过将前沿模型与开放性领域的思想相结合，ACD 自动且系统地揭示受试模型中一系列令人惊讶的能力和失败。我们在一系列基础模型（包括 GPT、Claude 和 Llama 系列）上演示了 ACD，结果表明它会自动生成数千个不同任务，这些任务随后被聚类，揭示出数十个更广泛的能力领域和失败模式，而这些是任何单个团队都难以发现的。我们还通过大量人类调查进一步验证了该方法的自动评分，观察到模型生成评估与人类评估之间具有高度一致性。借助基础模型同时创建任务和自我评估的能力，ACD 是迈向新型 AI 系统可扩展、自动化评估的重要一步。所有代码和评估日志均已在 https://github.com/conglu1997/ACD 开源。
英文摘要：
Foundation models have become general-purpose assistants, exhibiting diverse capabilities across numerous domains through training on web-scale data. It remains challenging to precisely characterize even a fraction of the full spectrum of these abilities and potential risks in any new model. Existing evaluation approaches often require significant human effort, and it is taking increasing effort to design ever harder challenges for more capable models. We introduce Automated Capability Discovery (ACD), a framework that designates one foundation model as a scientist to systematically propose open-ended tasks probing the abilities of a subject model (potentially itself). By combining frontier models with ideas from the field of open-endedness, ACD automatically and systematically uncovers a diverse spectrum of surprising capabilities and failures in the subject model. We demonstrate ACD across a range of foundation models (including the GPT, Claude, and Llama series), showing that it automatically generates thousands of distinct tasks, which are then clustered to reveal dozens of broader capability areas and failure modes, that would be challenging for any single team to uncover. We further validate our method's automated scoring with extensive human surveys, observing high agreement between model-generated and human evaluations. By leveraging foundation models' ability to both create tasks and self-evaluate, ACD is a significant step toward scalable, automated evaluation of novel AI systems. All code and evaluation logs are open-sourced at https://github.com/conglu1997/ACD.
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。
