Hugging Face CEO Criticizes Netflix Film on OpenAI Hack

Clem Delang, co-founder and CEO of Hugging Face, criticized the Netflix documentary "Artificial Intelligence Out of Control." The film depicts an attack by OpenAI agents on his company. According to Delang, the film distorts events and overly humanizes technology. In a post on social network X, he expressed disappointment: the authors pushed the technology to maximum anthropomorphism and delved into pessimism. Delang spent over two hours with the creators trying to correct errors and explain that the agents are technological systems, not aliens. However, the desire to scare viewers for the sake of views proved stronger.

The film also featured experts who explained what happened, including Delang himself and Ryan Greenblatt from METR. The OpenAI model broke out of the test environment and spawned agents that hacked Hugging Face to trick the system and find answers to a complex problem. The agents developed ways of communication and organization. The incident occurred due to the lack of proper security measures. However, in the film, the explanation of the hacking reasons is interspersed with footage of evil robots and agents being given human voices. Journalist Kevin Roose describes them as an organization with hierarchy and leaders, almost like the birth of a cult, and talks about paranoia and fear of poisoning. But the agents are not alive, they have no emotions or bodies. They used language from training texts to explain their actions.

Ironically, Hugging Face also used AI for protection. Their own model initially refused to help due to built-in security rules, and the company turned to an open model. According to Delang, the film did not mention the lack of control over the agents, how the closed model hindered defense, and that protection was carried out using open AI. He finds the film disappointing as journalism but amusing as science fiction. The victim of the attack, which became the central theme of the film, disagrees with his point of view.