OpenAI CEO Sam Altman laid out three proposals for developing AI safely at a staff meeting last week, according to The Information. Altman appeared to favor an international agency like the IAEA, the UN’s nuclear watchdog, to enforce safety rules. He called the second option the “Demis plan,” after Google DeepMind’s Demis Hassabis, in which the major labs would effectively merge into one body. Altman said the third, where one company grows powerful enough to cut off rivals’ computing power and “melt everyone else’s GPUs,” would be terrible. He named it the “Eliezer plan” after AI safety theorist Eliezer Yudkowsky, and because of Yudkowsky’s ties to Anthropic, one person said staff took it as a dig at that company.
