Gary Marcus on the "longgggg pause" when Altman tries to explain why anyone should trust OpenAI

Marcus on AI (Gary Marcus)·medium signalXBlueskyLinkedInCopy link

Writing September 16, Marcus attacks three positions at once. He contrasts Altman's current trust-us posture with his earlier admission that "You shouldn't" trust him, and cites the release of GPT-6 Astra despite it being less monitorable than its predecessor as proof safety loses to market share. He rejects Huang's corporate self-policing view ("I do wish what Huang was saying were true") and, unusually, also dismisses Bernie Sanders's claim that AI is more dangerous than nuclear weapons as sensationalism, noting nuclear estimates run to hundreds of millions dead. He redirects to Senators Hawley and Blumenthal, Asad Ramzanali on cybersecurity, and journalist Dan Primack.