DEBATE: A Large-Scale Benchmark for Evaluating Opinion Dynamics in Role-Playing LLM Agents

Name: DEBATE: A Large-Scale Benchmark for Evaluating Opinion Dynamics in Role-Playing LLM Agents
Uploaded: 2026-10-22T04:38:35.561282+00:00
Duration: 627 s
Description: DEBATE is a large-scale benchmark testing how well simulated opinions from role-playing LLM agents match real human interactions. It uses 2,788 participants across 107 topics to measure alignment at utterance, individual, and group levels. Results show that supervised fine-tuning (SFT) improves stance alignment and reduces over-convergence in the model's opinion dynamics.

DEBATE is a large-scale benchmark testing how well simulated opinions from role-playing LLM agents match real human interactions. It uses 2,788 participants across 107 topics to measure alignment at utterance, individual, and group levels. Results show that supervised fine-tuning (SFT) improves stance alignment and reduces over-convergence in the model's opinion dynamics.

Open the full episode