Synthetic Dialogue Generation with Role-Playing LLMs - Day 9 Morning

Learn to generate synthetic dialogues using role-playing large language models in this comprehensive tutorial from JSALT 2025. Explore fundamental concepts of how LLMs perform role-playing tasks, then delve into mechanistic interpretability techniques including difference of means, Sparse Autoencoders (SAEs), and the binding framework to develop actionable understanding of these models. Master dialogue generation approaches ranging from basic single-LLM setups to sophisticated persona-based multi-agent configurations, gaining practical skills in applying interpretability methods to analyze role-playing behavior in language models.