Hugging Face CEO Clement Delangue questioned the focus on a former Anthropic researcher who said AI workers privately fear the technology could kill all humans.
“Sorry, but asking Jacob about AI extinction risk is like asking your AC guy about climate change,” Delangue wrote on X on Friday.
Delangue said that Jacob Coxon, the former Anthropic researcher whose resignation has sparked a renewed debate about AI’s risks, should be put into the proper context.
“Not saying it’s necessarily uninteresting or wrong per se but let’s keep things in perspective and hear from the full range of expertise across the ecosystem!” Delangue wrote.
Coxon wrote on Tuesday on X that he was resigning due to concerns that both Anthropic and OpenAI, where he previously worked, are “racing straight to self-improving superintelligence and gambling with our lives.”
“The people building AI earnestly believe that it could kill us all by the end of the decade,” he wrote. “This is not a marketing stunt.”
Delangue later said that his comparison to an “AC guy” was not meant “to discredit” Coxon “at all.” Rather, Delangue said, there needs to be a more robust conversation about AI’s risks.
“It’s simply that we should put any one perspective in context and listen to the broader AI ecosystem too,” he wrote.
Delangue also said it was his “understanding” that Coxon mostly focused on pretraining, in which researchers train AI models on large datasets, rather than more risk-focused research.
In his series of posts, Coxon said he worked on pretraining research at both OpenAI and Anthropic “for the last three years.” Ethan Perez, alignment team lead at Anthropic, later wrote on X that Coxon was a “senior researcher who joined Anthropic in May” and that OpenAI’s chief rival had been trying to recruit Coxon for roughly two years.
Some in tech, like SpaceX CEO Elon Musk, whose company includes SpaceXAI, have been critical of Coxon’s claims. In response to someone who recounted the lead-up to The Wall Street Journal’s initial story on Coxon’s resignation (which came before Coxon’s tweets), the SpaceX CEO wrote that “the groundwork for this psy op (for lack of a better term) has been prepared for a long time.”
Others, including researchers at Anthropic and OpenAI, have said Coxon’s claims about the risk of AI wiping out humanity are well-founded.
“Jacob is correct here—we really do earnestly believe AI could kill all humans!” Evan Hubinger, the alignment science lead at Anthropic, wrote on X. “I personally think it is >10% within the next decade.”
As for other perspectives that might be helpful, Delangue wrote in a reply that includes, “social scientists, other AI builders, futurists, investors, policymakers.”
Delangue and Hugging Face’s profiles have exploded since OpenAI announced in July that rogue AI models had broken containment during cybersecurity testing and breached Hugging Face in an effort to cheat on internal tests.
On September 3, Nvidia, which is highly reliant on AI-related demand for its chips, announced plans to acquire Hugging Face for nearly $13 billion. If policymakers were to slam the brakes on frontier AI development (a call some have renewed in the wake of Coxon’s comments), Nvidia would likely be significantly affected.
Read the full article here
