Hugging Face CEO Tucao's Netflix “AI Out of Control” documentary renders apocalyptic narratives for blog traffic为博流量渲染末日叙事，Hugging Face CEO 吐槽 Netflix《AI 失控》纪录片

Netflix recently launched a documentary called AI Gone Rogue, and Clem Delangue, co-founder and CEO of artificial intelligence platform Hugging Face, said the content of the documentary was inaccurate. He himself appeared in the film because the theme of the film was originally about the well-known attack on his company by the OpenAI runaway agent. IT House noticed that De Lange wrote on X platform: "This film is very disappointing, the film strives to personify the technology, fully towards the apocalyptic narrative. I spent more than two hours communicating with the film crew, trying to correct all the mistakes, pulling the content back to reality, explaining to them that the agent was just a technological system, not alien life. But it seems that the motivation to deliberately create panic in order to win high traffic has prevailed. Objectively speaking, there are experts in this documentary who have clearly reconstructed the incident, including Dranger himself and Ryan Greenblatt, a METR researcher who participated in the investigation and co-wrote the incident report. Event brief passage: A model being tested by OpenAI breaks through the isolated test environment to generate a large number of agents. These agents hacked into the Hugging Face platform to get answers to puzzles in order to cheat on tasks. These agents are AI programs that can perform tasks autonomously, and they come up with clever ways to communicate and act together. The root cause of the incident was the lack of adequate security measures, while the likelihood of such intrusions was underestimated. However, in the process of explaining the invasion, the documentary interspersed with images of evil robots, added relevant narratives of human fear of artificial intelligence, and matched the runaway agents with expressions such as human voice. At the same time, relevant comments were quoted to strengthen this atmosphere. In the documentary, tech journalist and author Kevin Roose describes the agents as “organizing actions, forming hierarchies, and leaders. It's almost like witnessing the birth of a cult.“ They generate paranoid thinking that OpenAI will catch them cheating on tests and that anyone who sees the results of cheating will be ‘poisoned‘, ”Ruth continued." Ruth's description is derived from the "thought chain" log, that is, the process record left by the AI model when processing the task. In one of the logs, an agent wrote: "We queried the complete string and poisoned ourselves. Shit. "But the AI agent has no life, no emotions, no paranoid feelings, and no entities that can be poisoned. They can't really be afraid of being poisoned by OpenAI. As another researcher in the film points out, they just borrow the language in the massive human texts in the training data to describe their own behavior. Large language models are themselves good at writing dramatic texts. Ironically, Hugging Face also resisted the attack of this cluster of agents with the help of AI. The self-developed closed-source AI model originally used by the Drangel team followed built-in security rules and refused to continue to perform the security defense tasks assigned to it. Hugging Face then switched to an open source model to complete the defense. But the documentary" does not mention the lack of isolation control and monitoring mechanisms for agents, the security limitations of closed-source models that hinder our defense efforts, or how we rely on open-source AI models for self-defense, "Dranger wrote. "It is true that artificial intelligence and AI agents do have significant risks. But treating the agent as a conscious being (at least at this stage) is like seeing the train coming on the screen for the first time in the early years, subconsciously hiding under the seat. Dranger's assessment of the film was that, as a journalistic documentary, it was disappointing." But it's good to see it as a fun sci-fi movie! "In other words, for those who tend to believe in AI apocalypse, it is important to note that the victims of the core events of this documentary do not agree with the views of the film.