All the ways AI could wipe out human race - from nuclear Armageddon to paperclips - AOL

Concerns about AI’s rapid progress are being framed around scenarios that could threaten humanity, including warnings from researchers tied to major labs. The piece says AI researcher Jacob Coxon left his role at Anthropic after only four months, arguing he didn’t want to participate in efforts that could end civilization as currently understood. It reports Coxon said developers believe AI could kill everyone by the end of the decade, and that senior Anthropic staffer Evan Hubinger agreed the risk within the next decade is above 10%. The article discusses arguments about why humans may be unable to fully anticipate how a superintelligence could act, referencing Toby Walsh and a Nick Bostrom “paperclips” thought experiment, while noting debates over whether such a system would have the power to carry out goals.