arXiv 2312.147512023-12-22cs.LGcs.CY

日益可访问的可下载基础模型微调带来的危害

Hazards from Increasingly Accessible Fine-Tuning of Downloadable Foundation Models

Mila - Quebec AI Institute（魁北克人工智能研究所）；Université de Montréal（蒙特利尔大学）；UK AI Safety Institute（英国人工智能安全研究所）；University of Cambridge（剑桥大学）

Alan Chan, Ben Bucknall, Herbie Bradley, David Krueger

备注：: Accepted as a spotlight workshop paper at the Socially Responsible Language Modelling Research (SoLaR) workshop, held at NeurIPS 2023

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文论证可下载基础模型日益便捷的微调可能增加危害，通过分析降低计算成本与分摊成本的研究，指出其助长恶意使用并增加监管难度，强调亟需制定缓解措施。

摘要：

预训练基础模型权重的公开发布，即所谓的可下载访问，使得无需承担高昂的预训练成本即可进行微调。我们的工作认为，可下载模型日益便捷的微调可能会增加危害。首先，我们重点介绍旨在提高微调可访问性的研究。我们将讨论分为两类：A) 降低微调计算成本的研究，以及 B) 提高跨更多行为体分摊该成本能力的研究。其次，我们认为，日益便捷的微调方法可能通过助长恶意使用和使对具有潜在危险能力的模型的监管更加困难而增加危害。第三，我们讨论了潜在的缓解措施，以及更便捷微调带来的益处。鉴于危害方面仍存在相当大的不确定性，我们最后强调亟需制定缓解措施。

英文摘要：

Public release of the weights of pretrained foundation models, otherwise known as downloadable access \citep{solaiman_gradient_2023}, enables fine-tuning without the prohibitive expense of pretraining. Our work argues that increasingly accessible fine-tuning of downloadable models may increase hazards. First, we highlight research to improve the accessibility of fine-tuning. We split our discussion into research that A) reduces the computational cost of fine-tuning and B) improves the ability to share that cost across more actors. Second, we argue that increasingly accessible fine-tuning methods may increase hazard through facilitating malicious use and making oversight of models with potentially dangerous capabilities more difficult. Third, we discuss potential mitigatory measures, as well as benefits of more accessible fine-tuning. Given substantial remaining uncertainty about hazards, we conclude by emphasizing the urgent need for the development of mitigations.

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。