If we train a GPT model with our personal data to create a chatbot then will the chatbot only answer from trained data we have provided? or it will also answer on his previous data(pre-existing knowledge provided by GPT) that is provided by GPT already?
Both if I am not mistaken. It would just have your data as the freshest data, (maybe a better term for that) and it would most likely output data when it is specifically referring to your instructions becoming more relevant, I don’t think it would change much else. But that’s a lot of things that I have less experience with
Fine-tuning does not add new information to a model it teaches the model how to behave.
You will want to look at using an embeddings model and a vector database.