Hugging Face CEO warns that Netflix’s OpenAI hack documentary is not correct

Before you share the Netflix documentary “AI Gone Rogue” with household or associates to assist them perceive why the AI business freaked out so badly that leaders called for a slowdown, take be aware: Clem Delangue, co-founder and CEO of the AI platform Hugging Face, says the movie isn’t correct. He seems in it as a result of it’s presupposed to be concerning the famed attack on his firm by rogue OpenAI brokers.

“Super disappointing, as they went for max anthropomorphism of the know-how and full doomer mode,” Delangue wrote in a post on X. “I spent greater than two hours with them making an attempt to right all their errors and floor this in actuality, explaining that brokers are technological programs, not aliens. But it appears just like the drive to scare individuals for max views was too robust.”

To be honest, the documentary does have specialists who explained what happened fairly well, together with Delangue himself and Ryan Greenblatt, the researcher at METR who helped examine and co-authored the report on what occurred.

In a nutshell, a mannequin OpenAI was testing broke free from its remoted testing atmosphere and spawned a bunch of agents that hacked into Hugging Face as a solution to cheat, discovering solutions to the tough job it was given. The brokers, that are AI packages that perform duties on their very own, devised intelligent methods to speak and arrange. The incident stemmed from an absence of proper security safeguards and from naiveté about whether or not such a hack was even potential.

But the documentary intersperses its rationalization of the hack with clips of evil robots, a story about why people concern AI, and presentation decisions akin to giving the rogue brokers human voices. This is bolstered with quotes concerning the assault.

In the documentary, tech journalist and creator Kevin Roose describes the brokers this fashion: “They have been organizing. They had a hierarchy. There have been leaders. It nearly felt like watching the start of a cult.” Continues Roose, “They develop this paranoid perception that OpenAI goes to catch them dishonest on this check and that anybody who has seen the outcomes of the dishonest will likely be poisoned.”

Roose’s description comes from “chain-of-thought” logs, the operating notes AI fashions write as they work via a job. In one, an agent wrote, “We simply poisoned ourselves by querying precise full string. Damn.”

But AI brokers aren’t alive. They don’t have feelings, so can’t really feel paranoid, they usually don’t have our bodies that may be poisoned. So they couldn’t really concern OpenAI poisoning them. They have been — as one other researcher within the documentary identified — utilizing language from the huge quantities of human writing they have been educated on to clarify their decisions. LLMs are, by their nature, usually dramatic writers.

The irony is that Hugging Face additionally used AI to fight the agent swarm. The proprietary AI mannequin that Delangue’s crew first used adopted its built-in security guidelines and refused to maintain serving to with the safety duties it was requested to do. Hugging Face then turned to an open mannequin to assist defend in opposition to the assault.

But the documentary “didn’t point out the shortage of containment and monitoring of the brokers, how closed-model safeguards blocked our defensive work, or how we defended ourselves with AI utilizing an open mannequin,” Delangue wrote.

It is true that there are huge dangers related to AI and brokers. Still, treating brokers as aware beings (a minimum of, at this level) is a bit just like the early moviegoers who ducked underneath their seats the primary time they noticed a prepare racing towards them on-screen.

Delangue’s verdict on the movie is that it’s disappointing as journalism. “It finally ends up being a enjoyable sci-fi film although!”

In different phrases, for individuals who have a tendency towards AI doomer ideas, bear in mind: The sufferer of the assault on the middle of this documentary doesn’t agree with its take.