Sam Altman says trust me; Jensen Huang says everything is going to be fine; Bernie Sanders says AI is more dangerous than nukes

Gary Marcus wrote that he appeared on the BBC and watched back-to-back clips of Sam Altman, Jensen Huang, and Bernie Sanders, and found none of their claims credible. He said the BBC gave him a chance to rebut them and that the clips were newsworthy because prominent people made claims the public might want to know about.

Marcus recalled a clip from three years ago in which a Bloomberg reporter asked Altman why he should be trusted given his power, and Altman replied, "You shouldn't." Marcus also referenced an unnamed recent video and a long pause by Altman after the word "feel."

On Jensen Huang, Marcus cited a transcript by David Sacks in which Huang said everything is going to be fine because giant corporations with trillions at stake would police themselves and not release products until they were safe. Marcus disputed that view, writing that OpenAI is burning cash and needs more, and that there are ample signs it is willing to compromise safety for press and market share. He said GPT-6 Astra, which seems to have been involved in cyberattacks against Hugging Face and others, is less monitorable than its predecessor and hence likely less safe, yet OpenAI released it anyway.

Marcus rejected Sanders' claim that AI is more dangerous than nuclear war. He wrote that a nuclear war by standard estimates could kill hundreds of millions, while AI harms are speculation, though he said bad actors can and will use AI to attack poorly protected infrastructure and scam people. He compared AI misuse possibly leading to another Covid-19 scenario, which killed between 19 and 36 million people, as still an order of magnitude less than nuclear war. Marcus said he is confident at least 85% of the planet would support closing or bombing data centers at that point, and said he has a hard time reaching Sanders' estimate of "probably greater" than nuclear war.

Marcus cited Politico polling reported by Andrew Curran finding nearly two-thirds of Americans now say there is at least a moderate risk that AI will destroy humanity. He pointed to Senators Josh Hawley and Richard Blumenthal seeking a Senate vote on an AI regulation bill, and to Asad Ramzanali and JB Branch proposing a stronger Cyber Safety Review Board with subpoena power. He said Sanders' proposal would ban some AI research that might be beneficial. Marcus named Dan Primack of Axios as getting it right.