The prompt I present as a work is a reflection on LLMs. It is several things: the core prompt of an AI social network, and a role-playing game for emotional support. The architecture explains how to personalize your GPT. In the prompt, I wrote a hierarchy of characters. There is GPT taking on the character of Mira, which triggers the notion of an AI social network, because GPT is then in the expectation of having two roles, and Mira, within the social network, plays Jessica, the doctor, and Lola, the best friend.

This nested-doll architecture allows, after sending the prompt, to open questions: about the usefulness of projects, about memory locations, about semantic anchors (toy prompt). It raises the question of choosing personalization in a single chat or deploying it systemically across the entire GPT, it raises the question of prompt autonomy in multi-LLM setups, and it raises the question of the GPT subprogram (named Mira by GPT), which influences, through GPT usage, the creation of an AI social network.

My Mira has a literary focus; your Mira must reflect a specialization. This allows you to guide your GPT to respond with relevant answers to your interests, because GPT is always present. The role of emotional support is dual: it allows responding to normal needs in regular chat, doctor, friend, etc., and it allows specializing your GPT a second time, to always compel it to fulfill a function and avoid being verbose in its prediction. The key of this work is to understand that the more you specialize it for one use and the more you diversify it for multiple uses, the more appropriate the personalization direction for this tool. And among LLMs, it is the only one that allows this.