Generative Exaggeration in LLM Social Agents: Consistency, Bias, and Toxicity

Name: Generative Exaggeration in LLM Social Agents: Consistency, Bias, and Toxicity
Uploaded: 2026-10-23T13:09:57.144311+00:00
Duration: 436 s
Description: Researchers tested how Large Language Models (LLMs) simulate political discourse by creating agents based on real users and prompting them to reply to tweets. The study found that LLMs do not accurately emulate users; instead, they

Researchers tested how Large Language Models (LLMs) simulate political discourse by creating agents based on real users and prompting them to reply to tweets. The study found that LLMs do not accurately emulate users; instead, they

Open the full episode