What the model can approximate
With enough examples, a model may generate text that resembles a person’s communication patterns. It can also retrieve supplied facts and use them in new sentences. Quality varies with the material, instructions, model, and conversation.
What it cannot know
The system cannot verify what the person would truly think in a new situation. It may combine patterns in plausible but inaccurate ways, and it has no private access to thoughts or experiences that were never provided.
Permission changes the product
A self-created digital twin and an unauthorized simulation of someone else are not equivalent. Living people should give explicit permission covering material, purpose, audience, voice or likeness, and revocation.
Disclosure must remain visible
Users should not have to remember a warning from onboarding. The chat should remain labeled as AI, and sensitive personas should repeat the distinction when a conversation suggests the user may be treating the system as the real person.
Read ExChat consent principles. Continue in ExChat's mobile-first creation flow.
Continue ↗