Hi. As far as I’m understandanding, the answer to your question is that ChatGPT is based on a previous GPT-3 model, I guess text-da-vinci-003
.
Many (e.g. @daveshapautomator ) state that ChatGPT is just
1- a fine-tuned model of a previous GPT3 model (I guess the mentioned update of original davinci)
2- where the UX pattern is the question answering (pseudo chatbot) interface with simple web interface
3- temporary for free
BTW, The interesting part, for me is the meaning of ‘fine-tuning’ (possibly misleading). Don’t get me wrong: Ithink ChatGPT is maybe not a fine-tuned model, in the sense of the openAI API fine-tuning feature, but is maybe a special kind of hidden pre-emptive prompt engineering, where a previous GPT3 model is ‘programmed’ to act with a specific ‘bot-pesona’ (be polite, be complete and authoritative, etc.); so that persona is maybe a ‘tutor/teacher’…
Back to the original question of this popst: it doesn’t exist a ChatGPT API, even if you can workaround with some smart third-party SDks (e.g. in Python); see my post:
But guess what? You don’t need ChatGPT APIs, just because ChatGPT is a subversion/instance of a previous GPT3 model, for a specific use case (a playgrod to be used by the masses)…!
I experimented in facts that you can obtain BETTER results using official "old models’ APIs, where you tweak macro-parameters (temperature, etc.)
I hope this helps.