That would require the model to have the ability to attribute what training data sources it’s using to respond to a specific prompt, which AFAIK isn’t really co

That would require the model to have the ability to attribute what training data sources it’s using to respond to a specific prompt, which AFAIK isn’t really compatible with how LLMs work.