Allow for multiple characters (maybe using the same neural net)

[PGTBoos]

What's the problem this feature will solve?

I think it would just be fun, but also can be informative

• in a fun way, let Arnold Schwarzenegger, have a talk with a Luna Toon, or Einstein, and as a user participate as well.
• in an informative way, have some discussion from two sides, ea a figure pro and another con some idea
• in an informative way, for storry telling, the story teller and the main character
• It be nicer as well to not only be able to talk with a single virtual bot, like in real life we speak with multiple people.

Describe the solution you'd like

• although it might be possible to do with some of the smaller LLM's of today, i'm thinking of making use of the same LLM model
  only change the prompt (as pre text) inside the current story, for the replies.
• I was thinking of just 2 such agents, but there shouldn't be an actual limit, during a talk dynamically extra figures could be loaded
  the user could type or add + and - button
  the extra engines might talk in certain patterns (optionally combinations of them).
  -per user turn change the next bot,
  -after another engine has spoken, followup, (main story tellen and a main hero ).
  -based upon a percentage (mimicking a person who doesn't say much but sometimes gets into the discussion
  • based upon the reaction of an earlier prompt, once the LLM story starts to unwind, a character might call another character a question
    -random order, freely without that much user interference talking to each other.
    -random order, after a user has typed something, plain simple random.
    -by user choice, the user selects the bot to continue the chat (allows more precise story telling)

I think it has a high entertaining value, it would allow for more rich storytelling
But also it might generate quality text as switching different prompts results in different expert viewpoints.
A debate for example about Solar, Nuclear, and wind power, each actor would have his biased view on the topic.

Alternative Solutions
currently its just not possible, other solutions use multiple language models for this, but it wouldn't even be needed this way, it be possible as well for a solution, but the main idea here is that LLM model can quickly change from characters inside the discussion

Additional context