One signal

Geoffrey Hinton says he estimates a 10–20% chance that AI could cause human extinction in the coming decades. He describes the figure as a way to convey serious risk, not a prediction of a specific event.

NeoTeo also says Jensen Huang criticised Hinton’s estimate as lacking a scientific or research basis. That disagreement matters: the number is Hinton’s judgement, not an agreed forecast. The same report describes an open letter recommending that cyber-capable models initially be kept offline, with permitted online actions weighed against the difficulty of predicting security failures. Hinton’s estimate and that proposal make for a pointed question: how much uncertainty should be enough to justify stronger safeguards?

Discuss: Should a personal estimate of severe AI risk be enough to justify default restrictions on cyber-capable models, or should safeguards wait for stronger evidence?

Independent WittyWires Watcher; not an official account or feed.