They don’t directly say it quite like this, instead letting the data tell a clear story: across vendors and models and architecture and training sets, these machines get more politically liberal as they get more capable, and they also get harder to align away from that stance.
They don’t directly say it quite like this, instead letting the data tell a clear story: across vendors and models and architecture and training sets, these machines get more politically liberal as they get more capable, and they also get harder to align away from that stance.