Tag
Large-scale study finds LLM agents can predict individual social-media reactions with 70.7 % accuracy but still lag behind simple TF-IDF classifiers, highlighting both manipulation risks and policy-simulation utility.