Public framing
Persona is not just a system prompt. Long-term character agents need voice, memory, values, relationship state, and behavioral boundaries.
This lab explores how to reduce character drift over longer interactions without relying on private chats, copyrighted characters, or vague prompt-only definitions.
What it is for
- Modeling stable voice and behavior.
- Separating memory from personality.
- Making relationship state explicit.
- Testing boundaries and consistency over time.
Safety boundaries
- Use original sample personas for public demos.
- Do not publish private conversations.
- Do not lead professional framing with AI romance.
- Do not use copyrighted character material in public examples.
Questions
- What belongs in persona, memory, values, and relationship policy?
- How can the system notice drift?
- How should a character change without becoming incoherent?