So they want mathematicians to keep prompting and giving feedback to the models for clarifications on their proofs. Then write it all up and publish it in a journal and mention their AI usage. Then suddenly the models end up producing readable and understandable proofs through the "emergence" of the skills because of the scaling laws.