How does Copy.ai use GPT-3 and fine-tuned models for its AI writing platform?

When somebody's using Copy.ai, we track things like whether they copied the text, saved it, or did anything else with the results, because that shows signal that what we produced might actually have been useful. 

We take all of that data—whether they tried to rewrite it, save it, or whatever—and we  use a lot of that data to basically retrain a new model that has that same input pairs, but with data that we know humans like. 

That then results in a new model that is more human friendly.