Interesting idea.
I assume the agents are trained on data that was collected with respect to the persona of the AI agent you are trying to mimic.
I wonder how much better/worse this is compared to generic discussions with an LLM, or, even just simple google searches to answer questions.
Also, I wonder if the agents can address recent events as usually, LLM's are limited to information up until their training date.
For example, a stock market persona would have to be updated with the current market events.
Anyway, I liked the idea.
