Study Proposes Method to Measure Model-Specific Alignment Drift in LLM Opinion Agents
A new arXiv paper introduces a way to measure alignment drift in LLM agents used as synthetic participants in opinion-dynamics experiments, where agent behavior is often assumed stable.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-04 05:00 · arXiv cs.AI
Measuring Model-Specific Alignment Drift in LLM-Based Opinion Dynamics