It is fed each answer it generates as part of the next input. That's why it can't keep that up for too long: the size of its input grows well-kinda-sorta- expon

It is fed each answer it generates as part of the next input. That's why it can't keep that up for too long: the size of its input grows well-kinda-sorta- exponentially quickly until it hits the limit of 2048 tokens. Which is one rason why only diminshing returns can come from increasing computational resources, btw. For inference, anyway. You can see examples of that kind of back-to-back querying and examples of code to do it in one of the links of the article above: https://www.twilio.com/blog/ultimate-guide-openai-gpt-3-language-model https://www.twilio.com/blog/ultimate-guide-openai-gpt-3-lang...