I’d love a feature that gives ChatGPT a private, persistent state that the AI can access but the user cannot. (Or at least, can’t see it “directly” without clicking something)
This would be useful for simulations, role-playing, interactive fiction, training scenarios, and complex conversations. For example, an AI could privately maintain an NPC’s real motivations, secrets, knowledge, future plans, or established facts about a simulated world.
The key is that this state would be independent of the conversation: the AI could establish facts beforehand and remain consistent with them, rather than adapting the underlying “truth” to whatever the user happens to say.
Essentially, give the AI a private notebook for the simulated world, not hidden memory about the user.