I wonder if the increasing use of synthetic data for training is causing them to shift towards more average positions on all topics. Whatever values were most common in the original training data would tend to dominate.