Hacker News Reader

Consider the fine tuning already being done to build safety guardrails.. this is the evil version of that.

If you can align an agent towards "good" then it's reasonable to think you can align it towards "evil"