Aligning Large Language Models with Diverse Political Viewpoints

Large language models such as ChatGPT often exhibit striking political biases. If users query them about political information, they might take a normative stance and reinforce such biases. To overcome this, we align LLMs with diverse political viewpoints from 100,000 comments written by candidates running for national parliament in Switzerland. Such aligned models are able to generate more accurate political viewpoints from Swiss parties compared to commercial models such as ChatGPT. We also propose a procedure to generate balanced overviews from multiple viewpoints using such models.

Media Slant is Contagious

This paper analyzes the influence of partisan content from national cable TV news on local reporting in U.S. newspapers. We provide a new machine-learning-based measure of cable news slant, trained on a corpus of 40K transcribed TV episodes from Fox News Channel (FNC), CNN, and MSNBC (2005-2008). Applying the method to a corpus of 24M local newspaper articles, we find that in response to an exogenous increase in local viewership of FNC relative to CNN/MSNBC, local newspaper articles become more similar to FNC transcripts (and vice versa). Consistent with newspapers responding to changes in reader preferences, we see a shift in the framing of local news coverage rather than just direct borrowing of cable news content. Further, cable news slant polarizes local news content: right-leaning newspapers tend to adopt right-wing FNC language, while left-leaning newspapers tend to become more left-wing. Media slant is contagious.