habryka comments on [AN #115]: AI safety research problems in the AI-GA framework