this is some freaky stuff... what r the implications for language models like chatgpt? https://www.nature.com/articles/s41586-026-10319-8