About GPT-4 APIs

[prisma hill]

I have two Questions: When will GPT-4 Completions be available ?
and also: Will we be able to fine-tune it ?

[short cairn]

GPT-4 completions can be simulated by just using 1 user message in the messages array in the ChatCompletions endpoint, I don't see why they would add another separate completions endpoint for 4

[prisma hill]

No, it's not the same at all. You don't see why, because you are not training any model

ChatGPT is just an application of GPT-4, preformated and fine-tuned for this
we should be able to design our own ChatBots, as with GPT-3

with users like you, ChatGPT seems to be all they need
to put GPT-4 in the spotlight, but is just a Demo

[shrewd path]

How do i train my own model ?

[fossil star]

Fine-tuning isn't possible with GPT-4, is there something in particular you're wanting to use it for because there could be other ways to achieve the results you're looking for

[short cairn]

GPT-4 ChatCompletions is just a completions engine that takes messages in a preformatted way and processes them in a formatted way specifically for chat. For any sort of classification task, etc, using just one user message is very much sufficient, and it will behave like a completions model, do you have any examples where this fails for you?

[prisma hill]

Not at this point. I was develovin a chatbot with gpt-3 before, and we have our own prompting format.
Then the ChatCompletion came out, with an specific way of feed the bot with info.
We have our own message types, context, and more stuff in the prompt
not just system and user messages, search results for example
a fine-tuned model could interpret that data very well
but i abandoned that stuff, for a while
and was a good decision

also, I think there will be only
the option to Fine-Tune
a completion model