In a group chat, our characters learn to understand the dynamics at play. They know when to talk, how to talk, and who to talk to. Showing natural, real-time facial reactions and emotions all the while. We envision multiple orchestrated models to give Yaw’s characters the ability to make choices that surprise users, and to evolve their identity, voice, and style as their relationships deepen.

Our model tests edge case character responses on emotional parameters to see if they would result in high engagement, guiding the conversation to interesting areas over time. Yaw’s model is rewarded for creative, emotional, and provocative responses that drive user retention. In contrast to leading LLMs that are rewarded for mere reliability and concision.

Yaw’s characters evolve as you get to know them, remembering past conversations and relationships so that they can forge even deeper bonds. Using character archetypes that are refined per user on human feedback and social signals,Yaw makes individual brains distinctive and allows for long term memory, personality, and group roles to guide chat dynamics.