MIT built a dashboard to watch chatbots answer election questions in real time

MIT researchers built the LLM Election Observatory, a public dashboard that runs the same fixed set of midterm questions through popular chatbots every few days, varying the simulated voter's location, party affiliation and identity.

The team — Adam Berinsky, Chara Podimata and Charles Stewart III — wants to measure whether models answer the same question differently, whether answers drift as the election nears, and which sources the bots pull from.