Generative AI and elections

LLMs are increasingly intervening on the information ecosystem via chatbots, LLM-written news articles, and AI “slop”. Building on recent work, there are several directions of interest:

1. How is generative AI being used to generate personalized, persuasive content?
2. How aligned were LLM responses with Google search information?
3. To what degree did LLMs propagate misinformation during the 2024 election?
4. Did models change their behaviors in response to key election events in 2024?
5. Are models capable of understanding and reflecting on their own behaviors?
6. What is the minimal set of responses one needs to reliably predict a model’s behavior, i.e., responses to other queries?

These projects will utilize a variety of skills, from data analysis to causal inference. As a starting point, we may build on recent work and the corresponding dataset here. The goal is to publish a paper and policy report by the end of the year, aiming for a top ML venue.