Debian开发者josch公开声明，除非提供退出选项，否则将不再向NEW队列上传任何软件包 1。这一决定源于他发现DFSG团队在未发布公告的情况下，开始使用基于大型语言模型（LLM）的工作流处理NEW队列中的上传包 1。josch批评此举违反了Debian关于生成式AI使用的共识决议（GR），认为该行动构成了具有广泛项目影响的自动化操作，却未寻求事先讨论与共识 1。他援引GR决议文本指出，虽然Debian既不背书也不禁止使用生成式AI，但意图执行对全体项目产生重大影响的贡献者应当先寻求讨论和共识 1。
据josch透露，相关审查结果明确标注为“这是使用Debian推理工作流的基于LLM的审查。它不是由DFSG团队成员撰写的，也没有经过他们的检查” 1。面对这一现状，josch提出了两个选择：要么花费时间阅读可能错误的机器输出，要么浪费资源去生成这些输出 1。他建议系统默认设置应为“无LLM审查”，要求开发者必须显式选择加入（opt-in）才能使用该功能 1。此外，josch强调LLM本质上是一个不可复现、无法修复漏洞的“二进制黑盒”，相比之下，像Lintian这样的自由软件工具更具可复现性且值得优先优化 1。
Debian developer josch has announced that he will stop uploading packages to the NEW queue unless a mechanism is provided to opt out of automated reviews conducted by large language models 1. This decision follows the discovery that the DFSG team had begun using an LLM-based workflow to process submissions without issuing any public announcement 1. The developer explicitly stated, "I will no longer upload to NEW unless there is a way for me to opt out of this," citing concerns over the lack of transparency and consensus in the implementation of these tools 1.
Josch argues that the introduction of such automation violates Debian's General Resolution (GR) on generative AI. He references the resolution text stating that while "Debian neither endorses nor prohibits the use of generative AI," contributors intending to perform actions with broad project impact "should seek prior discussion and consensus" 1. He criticizes the current setup as opaque, noting that LLMs are effectively "binary blobs" that cannot be reproduced or debugged, unlike established free software tools such as Lintian 1.
The review outputs are clearly marked with a disclaimer: "This is an LLM based review using a Debian Inference based workflow. It was not written by a member of the DFSG team and has not been checked by one." 1 Josch argues that developers face two undesirable choices under the current system: spending time reading potentially erroneous machine-generated text or wasting resources generating it 1. He proposes that the default setting should be "no LLM review," requiring developers to explicitly choose to participate (opt-in) rather than being automatically subjected to the process 1.
评论
还没有评论，欢迎留下第一条。