Gary Marcus：生成式 AI 无法可靠遵循简单通用指令
精选理由
Gary Marcus 转发并吐槽了一件事：给 AI 写条“别伤害人”的规则它都执行不稳，他说这个被低估的风险早晚会出事。
Gary Marcus 在 X 上引用 Matthew Zeitlin 的推文并评论称，生成式 AI 的核心难题在于它无法被信任去遵循“简单、可泛化的指令”。他举例说，即使写入“不要伤害人”和“总是服从用户指令”这类规则，模型也未必稳定执行。Marcus 认为多数人没有意识到这一点，可能给社会带来实际风险。该推文获得了 23 个赞和 7 次转发。
原文 · Gary Marcus
i don’t know whether this is a joke or serious but the whole challenge with generative AI is that it literally cannot be trusted to follow “simple, generalizable instructions”. the inability of most people to grasp this is going to bite society in the ass. Matthew Zeitlin @MattZeitlin this is so true, what if we just encoded AI with simple, generalizable instructions like "don't harm people" and "always obey what people tell you to do" 🔗 View Quoted Tweet 💬 8 🔄 0 ❤️ 23 👀 1771 📊 7 ⚡