First, we fine-tune our neural network using vectors from the publications of Paul Graham, Naval Ravikant, Elon Musk, Ben Horowitz, etc., and then we allow users to ask a question and select whose opinions to consider in the response.
Another reason to use embeddings instead of finetuning is the ability to switch between states, determining which data to use and which not.
Perhaps in the near future, I will describe in more detail what embeddings mean and how to work with them. Is that needed?
