I've understood that the GPT-3 language generation is "programmed" by the end user with a contextual language.
This language is structured by human perspective on reality and populated (illustrated) by the collective human subconscious.
The context must be in terms of the human subconscious understanding of reality.
This means that the trick is to collaborate with the machine with the composition of a document that is valuable to your ends.
One friend, for example, used it to trace a strategy when talking with other business.
My other friend is conducting consultation sessions with people who want to consult with the GPT-3 oracle.
Super interesting stuff. Very valuable.