I’ve done a bit of coding today and it seems that there are signifcant differences betwen the web version of ChatGPT (https://chat.openai.com/) and the ChatCompletion API.
- The token limit in the web is 4096 for a single side of the conversation. In the API it’s 4096 combined. Half!
- The way the AI follows instructions and formats it’s answers is completely different.
I don’t mind the fact there are significant differnces but are they documented anywhere?
I was under the false impression the gpt-3.5-turbo model is what powers the web version of the chat.