评论：开源模型安全审计或成 OpenAI 与 Anthropic 的镜鉴
精选理由
这条推文聊了个有意思的角度：开源模型被解除对齐后，反而可能逼大厂正视安全问题，看法挺少见的。
一篇 X 帖子讨论开源模型的舆论走向。作者认为，若 OpenAI 与 Anthropic 拒绝或无法为所有软目标提供安全加固，用户将被迫用 abliterated 1017 这类去除安全对齐的模型自行审计系统。作者称这一局面会让人认识到忠实 AGI（loyal AGI）的必要性。
原文 · Teortaxes
Probably the best outcome for open model optics rn is if OpenAI and Anthropic refuse to shoulder the burden of hardening all soft targets (or can't), and people are forced to audit their own systems with abliterated 1017's, making it clear that loyal AGI is a necessity.
