search

ben-aaron188 / rgpt3

Making requests from R to the GPT models
GNU General Public License v3.0
101 stars 16 forks source link

Maintaining session/context? #14

Closed ahdee closed 1 year ago

ahdee commented 1 year ago

Hey thanks for this package. I have 2 questions:

  1. I think this is an API limitation but I'm wondering is there a way to maintain the state/context for each API call, like how the actualy chatgpt can keep each session so that we don't have to repeat what was prompted previously? Basically maintaining historical context.
  2. I notice that for each function call there is a model parameter; I'm assuming we can just input something like gpt4 or whatever model and it should work?

thanks.

ben-aaron188 commented 1 year ago

Hi. The my answers below:

  1. Yes, this is due to how OpenAI allows interaction with the model via the API. If you chat to ChatGPT today and tell it a few facts, then it won't "remember" these if you start a new session a week later. You can maintain a "within-session" context though the same way you can ask ChatGPT about previous answers in the session. This is something I've not yet had the time to test and implement.
  2. In principle: yes. GPT-4 (and it's model parameter counterpart) may come with some additional intricacies (as was the case for the GPT-3.5 to ChatGPT transition). GPT-4 support will come soon.