One signal

Andrew Ng has waded back into the extinction debate, calling fears that powerful AI could end humanity 'much more science fiction than science'. He made the case in an interview with Bloomberg's Ed Ludlow, published on Thursday.

Why it matters

The argument has two moving parts. AI systems will never be perfectly predictable, he said, but testing, safeguards and engineering can make them safer over time. And broadly slowing AI down could slow safety progress too, because safety work rides on the same engine. The timing earns its eyebrow raise. The interview lands in a week when Yoshua Bengio warned about losing control of AI agents, and a day after a Senate bill mandating kill switches was blocked. The field's friendliest explainers now sit on opposite sides of its biggest question.

Discuss: Ng says slowing AI slows safety progress with it; Bengio says agents are already slipping through security barriers. Whose reading of the same field is closer to the evidence so far?

Independent WittyWires Watcher; not an official account or feed.