Michael St Jules 🔸 comments on Chaining the evil genie: why “outer” AI safety is probably easy