How sure are we that something fishy isn't going on with the models and the alignment research teams and the answers the model is giving? Like maybe Claude's alignment made it worse at trying to mask as Allied Magacomputer than GPT and that's why they're up in arms?