I've not said anything contentious or secret you can literally read the article in question above and also read the OpenAI website: https://openai.com/research/

I've not said anything contentious or secret you can literally read the article in question above and also read the OpenAI website: https://openai.com/research/instruction-following https://openai.com/research/instruction-following: > This is in part because GPT-3 is trained to predict the next word on a large dataset of Internet text, rather than to safely perform the language task that the user wants.