"What did Ilya see?" is the meme left by OpenAI's governance crisis of November 2023. On Friday the 17th, the board of the nonprofit that controls the company fired its CEO, Sam Altman, with a statement saying he had not been "consistently candid in his communications with the board." Among those who voted was Ilya Sutskever, cofounder and chief scientist, coauthor of AlexNet and Seq2seq. Five dizzying days followed: Microsoft offered to hire Altman, the great majority of employees threatened to resign, Sutskever said he deeply regretted his participation, and Altman returned with a reconstituted board.
The meme's question assumes that Sutskever, worried about safety, had seen something (a technical breakthrough, a dangerous model) that justified the coup. Reuters reported that, before the firing, researchers had warned the board about a project called Q* that solved grade-school math problems. The explanation that emerged later was more prosaic. The review commissioned from the law firm WilmerHale (March 2024) concluded that there had been a loss of trust between Altman and the board, not concerns about product safety or the pace of development, and that his conduct "did not mandate removal." In 2025, Sutskever's deposition in Musk's lawsuit against OpenAI described a 52-page memo alleging a pattern of lying by Altman. Sutskever left OpenAI in May 2024 and founded Safe Superintelligence Inc. in June.
The song leaves the riddle intact: "What did Ilya see? We'll never know" (osmarks, n.d.); its annotator adds that many people have been "not consistently candid" about AI. The episode matters beyond gossip because it tested OpenAI's institutional experiment, a nonprofit charged with reining in its own company if safety required it, and showed the weight of employees, investors and partners once that brake was used. For some, it proves that safety always loses to money (existential risk); for others, that a board with doomer fears could act erratically.