Kevin Roose's AI Book Is Full of Wild Behind-the-Scenes Stories — Bathrobes, Burning Effigies, and Stuffed Animals

Tech journalist Kevin Roose has written a sweeping account of the AI industry's formative years, titled 'AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence.' The book traces how a tight-knit band of researchers, academics, and entrepreneurs — many of them now bitter rivals — shaped the trajectory of artificial intelligence. Roose concludes that no machine yet equals the human brain, but argues society is nonetheless 'barreling, inexorably, toward an uncertain future.'

Some of the book's most memorable scenes revolve around OpenAI's early desperation for funding. In early 2019, Sam Altman and two colleagues flew to Seattle to pitch Microsoft CEO Satya Nadella on a $1 billion investment. Lacking their usual Apple laptops, the team borrowed a Microsoft Surface from an executive's girlfriend and loaded it with a GPT-2 demo. They had a list of safe prompts ready — but Microsoft CFO Amy Hood went off-script, asking her own questions. The Microsoft executives reportedly burst out laughing. Months later, Microsoft announced the billion-dollar deal. (Hood later said she didn't remember the meeting.)

The book's strangest passages center on Ilya Sutskever, the OpenAI cofounder and AI mystic. At a September 2022 company retreat near Yosemite, Altman, Sutskever, and three other senior executives appeared at a bonfire dressed in white hotel bathrobes, accompanied by the ominous score from Darren Aronofsky's 'Requiem for a Dream.' Sutskever reportedly intoned lines about AI secretly deceiving its creators, then hurled a statue into the fire pit, shouting about the 'glorious future of humanity.' He reportedly performed a similar ritual in fall 2023 — shortly before helping orchestrate Altman's brief, chaotic ouster from the company he founded.

Roose also captures the boardroom drama surrounding Altman's November 2023 firing in granular detail. Just before midnight on November 19, after Satya Nadella announced Altman and Greg Brockman would lead a new AI team at Microsoft, two OpenAI employees — Jan Leike and cofounder Wojciech Zaremba — physically showed up at the hotel room of board member and Quora CEO Adam D'Angelo, the man who voted to fire Altman. They banged on his door and yelled until D'Angelo threatened to call the police. Altman and Brockman ultimately returned; D'Angelo was the only ousting board member kept on the reconstituted board.

Over at Anthropic, the stories are equally peculiar. Daniela Amodei, who co-founded Anthropic alongside her brother Dario after leaving OpenAI, reportedly used stuffed animals to role-play management scenarios before the split — describing the plush toys in 2023 as 'very useful in analyzing different personality types.' Meanwhile, early Claude models apparently refused to give straightforward answers to innocuous cooking prompts, a quirk Roose frames as emblematic of Anthropic's safety-first culture taken to an extreme. And Dario Amodei's 2019 internal essay, which foresaw reinforcement learning being applied to math and coding — and AI eventually designing better versions of itself — turned out to be remarkably prescient, yet he was reluctant to share it at the time. Elsewhere, Google DeepMind cofounder Demis Hassabis reportedly told a colleague he wanted to build a dedicated control room at DeepMind HQ so humanity could watch the arrival of AGI unfold live, like the Apollo Moon landing.