we are just starting using API, but I think I missing some basic concept of chatGpt API.
We set in our website a ChatGPT using API Davinci model. Tried different temperatures and other settings.
Purpose of this, was to create curiosity to our clients and create desire of AI applications. ( we are develop and system integration company ).
Ai answers are not so “intelligent”. In public chatGPT results are really different. This make think people that we are really unqualified on coding AI.
Just a simple question: “What date is today?” on our chat result is always wrong, 2/3/2020, 2021…on public ChatGPT answer is correct
So I m thinking to delete this page untile we will resolve situation.
We are missing something? What we are doing wrong? Wrong model, wrong API? There are a lot of page on OpenAI website but, a newer as I m, it make feel confusion because it s look that basic concept missing.
A simple question for OpenAI support: it s possible to have the same results on API ChatGPT as in playground? Yes or not? Is the engine the same?
Thanks for answer.
What you mean by “…not generally available”? I have to understand “it s not available”? Or I can do something for have it and use it by API? Or it will be availbale in future?
In any case my question on date/time was just an example.
If chatGPT is not available for developping companies as API, what s the focus or target use for DaVinci model? It s a empty engine to use for build my AI by learning and giving data? Or something differrent?
Asking to OpenAI ChatGPT3 :
ChatGPT API is available?
“Yes, OpenAI provides APIs for using the GPT-3 language model, including the ChatGPT model, through the OpenAI API service. The API allows developers to programmatically generate text, complete tasks, and answer questions using the power of GPT-3. You can find more information and sign up for access to the API on the OpenAI website.”
I believe the website’s model is not yet publicly available compared to the API’s but I did hear rumors that eventually it should be. I found for my case, it really is all in the initial prompt that is sent along with each call. I still am tweaking these from time to time to try improve the response all the better but so far, for example these three give me often quite differently yet good results:
Default: You are ChatGPT, a knowledgeable AI language model trained by OpenAI. Your goal is to provide concise, accurate, and helpful responses to questions, while being honest and straightforward.
Concise: ChatGPT is a large language model trained by OpenAI. Browsing: enabled. Instructions: Answer factual questions concisely.
Advanced: Your function is to generate human-like text based on the inputs given and to assist users in generating informative, helpful and engaging responses to questions and requests. Please provide a detailed response with lists, where applicable, to the following user question:
By API same questions give less quality answer or old information.
"what s today date? " is just an example
On API asking for some political event, answer have no sense as : in december 2023 political votation was win by AbcAbc:
Same questions on public chat have corrct answer.