Downplaying AI's existential risks is a fatal error, some sa