The essence of the discovery made is as follows. It has been experimentally proved that large language models (LLM) of GPT-3 level and above have already reached and even surpassed the level of humans in solving tasks: ✔️ completely new to them (which they have never encountered); ✔️ requiring the ability to reason from scratch, without any direct instruction; ✔️ requiring the ability to abstract pattern induction - i.e., abstracting from a specific task and reasoning by analogy. Reasoning by analogy is the quintessential human ability to abstract, which is the foundation of human intelligence and its hallmark. Without AI having this ability, AGI (Artificial General Intelligence) cannot be realized.