API for the exact GPT-4 used in the ChatGPT GUI?

Is there a way to access GPT-4 in the version is it finetuned/prompted in the GUI version? I have a task for which using gpt-4-1106-preview performs fairly well (~95% accuracy) but asking GPT4 with the same user prompt in the ChatGPT interface produces correct results every time.

In case there should be no such endpoint:

  1. Do you have any advice on how to get gpt-4-1106-preview to behave as closely as possible to the GPT-4 in the GUI?
  2. If even that should not help, I would just get some training data from the GUI and finetune a Mixtral8x7B instance. But (assuming I have already reached the maximum potential of CoT prompting for my use case) do you think there is any simpler way of achieving my goal?

Thanks in advance!

Hi! Welcome to the forum!

I’m assuming with the GUI version you mean ChatGPT, and not the playground.

Unfortunately, as far as we’re aware, OpenAI injects some stuff into the prompt that we don’t know. It’s also possible, that ChatGPT runs on a model or finetune that isn’t available on the API.

The advice I would give is to not try to get it as close to ChatGPT as possible. My advice would be to refine your prompts and your parameters to get the outcome that creates business value for you.

If you wanna share details of what you’re trying to accomplish I’m pretty sure that some folks on here would love to take a crack at it.