Their finding of LLMs working best at simple tasks, LRMs working best at medium complexity tasks, and then neither succeeding at actually complex tasks is good to know.