proving my point. the training set can be improved (until they’re irreversibly tainted with llm genrated data). The tech is not. Even with a huge dataset, llms will still have today’s limitations.