ai doomers' 10% extinction risk by 2030 may be overblown
anthropic and openai employees express fears of ai killing humans, but expert gary marcus says focus on end-of-world scenarios is misguided and harmful.
a recent thread by jacob coxon, a former employee at openai and anthropic, sparked controversy by suggesting that ai could kill humans within five years. coxon's claims were corroborated by evan hubinger, a current anthropic employee, who estimated a >10% risk of ai extinction within the next decade. however, gary marcus, an expert in ai, argues that the focus on these 'doomer' scenarios is misguided and has not led to demonstrated solutions to the core problem of ai safety. marcus claims that this narrative has instead enabled frontier ai companies to accumulate wealth and power, potentially increasing the risks associated with ai.