Huang replaces his pacing advice with a flat 0% on AI extinction risk

The earlier line was useful to somebody. A product manager writing the release gate on an agent feature needed a reason from outside the safety team, and a chip vendor conceding a brake was the most quotable reason available, since of everyone arguing about pace Huang has the most direct financial exposure to the answer ^[4]. That citation is spent.

Two sentences in the excerpt now have to live together. If pace is set irrespective of anyone else, readiness is judged by the shipper, and the only check on release is the releasing company's own assessment. TNW notes the pacing proposals were written to change that arrangement ^[9].

An outside quote wins a scheduling meeting; it does not set a threshold or write an eval. The gate underneath rests on numbers a team measured itself, like the eval a build failed before launch and the time a rollback actually took. Those numbers are unchanged, and the only thing lost is a sentence a PM could read aloud when the launch date was under pressure.

Zero is also a number that only Huang brings to this argument. The labs publish safety frameworks, capability thresholds and risk estimates. Those are instruments for operating under uncertainty ^[17]. The pacing case is explicitly a bet about unknown probabilities, which is why it asks for evaluators and tripwires instead of a halt ^[18]. Huang's zero is the only point probability on the table ^[19]. His earlier answer was narrower: he called apocalyptic predictions not grounded in science, which is a claim about evidence ^[3]. Andrew Ng has made that narrower case without any inventory to clear, calling the extinction warnings science fiction ^[16]. The people being dismissed include a DeepMind AGI safety researcher who left this month and published his warning ^[12].

Trump has spent the month attacking AI safety warnings and calling them a hoax, repeatedly and at length ^[10], and CBS trailed the interview on Huang agreeing with him ^[11]. Huang has been publicly resisting proposals to sanction Chinese AI ^[6]. He is expected at the state dinner for Xi Jinping next week ^[7].

The cheap test for anyone defending a gate is to cover the outside name on each line of the checklist. The lines that still give a reason point at something the team owns, like the threshold agreed before the build or a rollback time someone measured. The rest were borrowed from a supplier whose answer changed between two interviews ^[15].