That’s why, in the case of a GPT model you would feed it custom training data using something like LlamaIndex. I don’t know if there’s an API available for Bard, tho.

You’re wrong assuming that the free models that we have at our disposal are the only possible and best implementations of these LLMs.