That isn't what is being described here. They are just providing additional context to ChatGPT using its plugin API. It's still trained on large amounts of public text data.