I've recently read about them being principled about making sure not to train their models to be sneaky in ways they wouldn't be able to detect. That would be a good explanation for models not trying to _hide_ when they're being sneaky.
No comments yet.