Regression to the Mean: on LLMs and the quiet death of the new · HackerLangs