“The Effective Altruism community historically dismissed the risk of "malicious use" of AI, viewing it as a distraction from the primary risk of "loss of control."”