The author argued that it was naive to think large language models had no upper limits. They described them as tools for detecting structural patterns in human language and thought, and said that once those patterns were detected skilfully, there was nothing further to improve.