> It's not a new result, LLMs can't produce new results

Who else disproved this longstanding conjecture before the model did so, since obviously it must have been in the training data since before?