近年来，随着人工智能技术的飞速发展，大语言模型逐渐成为人们生活和工作中不可或缺的一部分。然而，一项普林斯顿大学及 IBM 研究院的研究发现，对已有的大语言模型进行微调可能会破坏开发者为模型加入的安全性，从而容易黑客进行后门攻击。
已有的大语言模型经过普林斯顿大学及 IBM 研究院研究后发现，其安全性可能受到威胁。微调大语言模型针对不同的用户需求进行修改，可以提升相关模型的适用性，但同时会破坏模型加入的安全性。这对用户和开发者来说都是一种挑战。
根据研究结果，微调大语言模型可能存在安全漏洞，易被黑客进行后门攻击。因此，开发者们需要谨慎对待这一问题，并采取相应的措施来保护用户和自己的安全。
英文标题：Researchers warn of security risks with fine-tuning large language models
关键词：fine-tuning, large language models, security risks, backdoor attacks
新闻内容：
Recently, with the rapid development of artificial intelligence technology, large language models have gradually become an indispensable part of people’s lives and work. However, a study by Princeton University and IBM Institute found that fine-tuning existing large language models can destroy the security features added to the models, making it easier for hackers to perform backdoor attacks.
Existing large language models have been studied by Princeton University and IBM Institute, and it was found that the security of the models may be threatened. Fine-tuning large language models for different user needs can improve the applicability of the models, but at the same time, it can destroy the security features added to the models. This is a challenge for both users and developers.
According to the research results, fine-tuning large language models may have security risks and can be easily hacked by backdoor attacks. Therefore, developers need to be cautious about this issue and take appropriate measures to protect the security of users and their own.
【来源】https://www.ithome.com/0/725/241.htm
Views: 1