Most Replayed Moment: How AI Companies Are Gaslighting You - Karen Hao

The discussion begins by comparing public statements with the incentives surrounding them. A 2017 statement attributed to Dario is presented: the chance of a civilization-scale catastrophe might be between 10% and 25%. The date and attribution matter. This is an individual's assessment presented in the conversation, not a measured probability or a consensus finding. The discussion then turns to Ilya's departure from OpenAI. Hao says her interviews led her to conclude that he saw Sam Altman's management as undermining both the pursuit of AGI and its safety; the excerpt presents her account of his concerns rather than a new direct statement from him.

Hao recalls interviewing Ilya in 2019 and reproduces his analogy between powerful AI and humanity's treatment of animals. A highway can destroy animal habitats without its builders hating animals. Within that analogy, the danger arises from unequal power and competing priorities rather than necessarily from malicious intent. Hao then identifies an assumption beneath this forecast: the view that brains are fundamentally statistical models and that sufficiently scaled statistical systems could reproduce or exceed human intelligence. She repeatedly calls this a contested hypothesis, not an established scientific conclusion.