AI safety isn't what people think it is. It is not "we need to prevent AI from becoming sentient and persecuting humanity", it is "how do we ensure that AI answers conform to what a small number of white american progressive elites think and not what most people think"