“Current large language models will fail to achieve robust reasoning comparable to a 10-year-old child by 2027.”