辛顿把今年夏天OpenAI智能体入侵Hugging Face平台事件，类比为核安全史上的三里岛事故，属于一次危险的预警信号，但这还不是真正的灾难。
所谓AI界的“切尔诺贝利”，并不是科幻电影里AI自主反叛、消灭人类，而是AI被黑客利用，发起针对关键基础设施的网络攻击，最典型的场景就是造成大范围长时间停电，或是让全球金融系统瘫痪一周，直接冲击民生运转秩序。
在采访中，辛顿直言自己并不看好头部AI企业管理层的风险态度，点名批评OpenAI的奥特曼、Meta的扎克伯格、马斯克等人。
辛顿表示，称他们清楚AI背后潜藏的巨大隐患，但在商业竞争和利益面前，没有给予足够的重视，各家只顾着比拼模型能力、抢占市场。
需要指出的是，辛顿并非反对AI，他认为AI就是一把双刃剑，自己认可其拥有巨大正向价值，能够给人类带来美好的未来，可一旦重大事故发生，人们或许会后悔当初不该发明这项技术。
在他看来，未来十年就是人类的关键窗口期，人类终将验证，到底能不能管住AI的风险。
那么该如何防范风险？辛顿给出的解决方案也十分明确：针对前沿大模型，必须落地强制安全测试、第三方独立评估，推动各国建立统一的AI安全标准，用监管约束狂奔的技术。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
The fact that Shinton compared this summer ' s invasion of the Hugging Face platform by OpenAIAI agent to the three-mile island accident in the history of nuclear safety was a dangerous warning signal, but it was not a real disaster.
The so-called “Chernobyl” of the AI world is not a science-fiction movie in which AI rebelled and eliminated human beings, but rather is used by hackers to launch cyberattacks against critical infrastructure. The most typical scenario is a massive blackout of power for a long time, or a week of paralysis of the global financial system, which directly affects the living order.
During the interview, Sinton stated that he did not appreciate the risk attitude of head AI corporate management, naming Ottmann of Open AI, Zuckerberg of Meta, Mask and others.
Sinton stated that they were aware of the enormous potential of AI, but did not pay sufficient attention to commercial competition and interests, and that they were focused on modeling and market grabs.
It should be noted that Sinton is not opposed to AI, which, in his view, is a double-edged sword that recognizes itself as having a great positive value that can bring humanity a better future, but one may regret that it should not have been invented in the first place if a major accident had occurred.
In his view, the next decade would be a critical window for humanity, which would eventually verify whether or not it could manage the risks of AI.
So how do we protect ourselves from risk? The solution offered by Sinton is also clear: for the front line of LLM, mandatory landing security tests, independent third-party assessments, and the promotion of uniform AI security standards in countries, with technologies for regulating the run.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.