I actually agree that stronger models are easier to achieve any given % alignment, but on the other hand the potential bad consequences of any given % misalignment increase for the stronger model (potentially dramatically at certain points, like it can take over).
I actually agree that stronger models are easier to achieve any given % alignment, but on the other hand the potential bad consequences of any given % misalignment increase for the stronger model (potentially dramatically at certain points, like it can take over).