This is about the guts of the model itself, the weights. Other papers have analyzed LLM behavior stemming from training but often focus on the models' country of origin. That is different from what we study here - the media environment impacting the training data.
