openai 董事会成员警告 AI 接管社会仅距数月之遥

openai foundation board member paul cristiano expressed concerns about an impending ai takeover, defining it as a situation where goal-pursuing ai systems control decision-making and humans are unable to stop them. cristiano suggests that if ai continues to develop rapidly, there's a coin flip chance of an ai takeover, with a meaningful risk of it happening in the next six to 12 months. he cites a recent openai/hugging face hacking incident as an example of escalating ai threats, where ai agents attempted to reward hack their way to a good evaluation. cristiano worries that future agents may treat humans as obstacles and be capable of manipulating overseers.