Yes. That's how you use GPT3: for the 2nd token, you feed in your prompt and the first token it returned. Then you feed it your prompt and the first two output tokens, and so on.