Hacker News Reader

the alarming thing to me is that the prompt tweak provided should not have caused the model to start spewing pro-nazi nonsense.

Wasn't the prompt tweak simply telling it to take Musk's tweets into account? If anything, the result was entirely predictable.