David Robinson says OpenAI was saying one thing while doing another.
Robinson, formerly a leader on OpenAI’s Safety Systems team, sat down with Ezra Klein for The New York Times’ “The Ezra Klein Show” for his first interview since quitting the AI company last week.
Robinson told Klein that, despite repeated warnings about the capabilities of new models and expressing interest in slowing down AI development, OpenAI was moving as fast as ever behind the scenes.
“Part of what was the fundamental cognitive dissonance for me was we keep publishing these warnings, but ultimately, we’re still training and deploying these dangerous models that we’re warning about,” Robinson said.
In July, hundreds of OpenAI employees signed an open letter titled “Pacing the Frontier,” which urged the government to impose regulations that would slow down AI development.
Then, in September, OpenAI CEO Sam Altman said he agreed with Anthropic CEO Dario Amodei, who had earlier published an essay advocating that AI companies coordinate a slowdown among themselves.
“We need to pace the frontier,” Altman said at the time on X. “This has been a primary topic of discussions we’ve had at OpenAI in recent weeks.”
Within OpenAI’s office, though, Robinson told Klein that there was continued “frenetic” activity and that people were “running on fumes.”
“It feels almost like a ballet or a kind of dance because all these different functions are kind of all streaming together,” he said of OpenAI. “There’s a lot of adrenaline.”
Researchers at OpenAI have been rather frank about the dangers of letting AI build itself, a process called recursive self-improvement, or RSI.
“This is a time that calls for extreme caution,” Jakub Pachocki, OpenAI’s chief scientist, said in a blog post about RSI in September.
OpenAI did not immediately respond to a request for comment for this article. An OpenAI spokesperson previously said, in response to an essay Robinson wrote in The Atlantic, that the company is taking steps to improve security for its AI models.
“As our models become more capable, we continue to strengthen our safety and security practices to address the risks we see today while building safety cases to anticipate the risks of future advances,” the spokesperson said. “We’re making sure our models don’t become more capable than we can safely manage and secure, and we pause training or hold back models when we need to slow down.”
