[Internal Discussion by Altman on Three Paths to AI Safety, the Most Extreme Being 'Melting Down Competitors' GPUs'] Beating AI Newsflash: OpenAI CEO Sam Altman introduced three approaches for safely developing superintelligent AI to employees last week. The first approach involves establishing a global regulatory organization similar to the International Atomic Energy Agency (IAEA) to inspect advanced models and audit safety measures. Sources familiar with the matter say Altman favors this route. The second approach is modeled after the European Organization for Nuclear Research (CERN), where major AI labs would merge into a single organization or collaborate through a unified institution for joint development, rather than continuing individual races. The third and most extreme approach, which Altman referred to as the 'Eliezer Plan,' involves one AI company achieving recursive self-improvement first, enabling AI to continuously enhance its own capabilities. This company would then leverage overwhelming economic power to control computational resources, preventing other developers from catching up. Altman even used the phrase 'melting down all other companies' GPUs' to describe this scenario but clarified that this plan is highly undesirable and was mentioned only to cover all possibilities. The idea of 'melting down GPUs' originates from AI safety researcher Eliezer Yudkowsky. As early as 2021, Yudkowsky proposed a thought experiment where a superintelligent AI could create self-replicating nanomachines to destroy GPUs worldwide, thereby halting other organizations from developing dangerous superintelligence. However, he explicitly stated that this was merely an extreme example to illustrate intervention measures and not a proposal he genuinely advocated. The first two approaches are also not new ideas. OpenAI proposed the establishment of an 'AI version of the IAEA' back in 2023, and Google DeepMind's head Demis Hassabis has previously advocated for creating an 'AI version of CERN.' Currently, OpenAI, Google, and Anthropic are planning to form a joint safety standards organization by the end of this year or early next year, but disagreements remain over who should be responsible for safety testing of the models. Altman hopes to set a 'speed limit' for AI development but also admits that OpenAI cannot afford to hand over its competitive advantage to rivals. [Original Link]