Does anyone with a better grasp on AI than me know: if folk are putting docs (for eg grant applications/papers) in chatGTP and similar to help improve text or even just reduce word count, does that doc then become part of the training data?