“The number one misuse of GPT-3 observed by OpenAI after its initial deployment was the generation of medical spam, an application the team had not anticipated.”