I 
love the completions from ChatGPT Plus/GPT4.
Is there a programmatic API that produce the same result as ChatGPT Plus/GPT4?
Why do I need this?
My experiments are in the classify+explain area.
The 0-shot and k-shot results from ChatGPT Plus/GPT4 are far superior to the normal completions APIs for GPT4.
The OpenAI GPT4 API over analyze and get the classification mostly wrong 
where as ChatGPT Plus/GPT4 gets it mostly right 
.
Any pointers?
Thank you.
The APi over analyzing could be a hint towards you using a higher temperature in your API call.
Try to play around a bit with the temperature, top-p and max tokens parameters of the API call, as they impact the quality and the content of your generation.
Temperature and top-p impact the tokens which are selected for the generation. A lower temperature means only words/tokens which are closely related to your topic will be selected.
Max tokens is the sum of tokens (input+output that the model can handle). Based around how many tokens are left for the output, the output could be variable ie more descriptive or short and sweet
The primary differences between ChatGPT and API are going to be system message and temperature (try 0.5-0.7).
Can you share an identical example to ChatGPT and the API?
thanks @novaphil @udm17 ,
here’s my setup for the GPT4 API
messages = [{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": prompt}]
Would you know how the ChatGPT/GPT4 is setup?
ChatGPT is set at a high (or default) temperature. It doesn’t produce the same output each time itself, by a wide margin.
system: You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture. Knowledge cutoff: 2021-09. Current date: 2023-09-09
(plugins enabled in ChatGPT is likely to have its own trained model API users can’t access.)
@_j , thank you for chiming in…
Are you suggesting that, to get ChatGPT Plus-like completions, I should be using the system value as you state above?
system: *You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture. Knowledge cutoff: 2021-09. Current date: 2023-09-09
ChatGPT plus, aka GPT-4, aka plugins, besides the system prompt, one would want to include a never-called function to get switched to a model with the same type of training on calling outside resources. The extra training on writing code makes plugin-activated AI have different skills and weights.
However, why would you want to make ChatGPT when that’s already a thing…
thanks…because there isn’t a programmatic interface that mimics ChatGPT Plus (and for my completions, ChatGPT Plus consistently does better)…which is all I want and don’t know how to get.