@Raemon777 This possibly is a reason why we have more time than we might expect before models become dangerous in the existential risk sense.
If a model if smart enough to quickly RSI, but will suffer from value drift doing so, it won't.
Fails in the limit case though.
@patio11 America gets a lot of things right, but UK261/EU261 when dealing with airlines is a good example of regulation forcing an industry to internalise the negative externalities of, in this case, flight delays.