I think the gpt-3 vs BERT story should be more interesting.
I was told that OpenAI practically tried to "label every human knowledge" prior to ChatGPT, sort of similar to what Li Feifei was doing.
At that time, no one seems to understand how to exact the massive capabilities out of gpt-3 (remember that gpt3 troll bot on 4chan?), until some joking threads on Twitter proposed "Lets think step by step..."
That seems to be the wakeup prompt to a new AI era.