They list several tests with record breaking results here: https://blog.openai.com/better-language-models/#zeroshot https://blog.openai.com/better-language-mode

They list several tests with record breaking results here: https://blog.openai.com/better-language-models/#zeroshot https://blog.openai.com/better-language-models/#zeroshot Are those previous records their own or were those the best scores in the whole field of NLP? It sure looks like a pretty big step forward for one model to break that many records.