Letting a GPTs talk to OpenAI GPT-4-Turbo | OpenAI | Page 1

pine mica Nov 13, 2023, 12:39 PM

#

Hey everyone, I have a question. In the OpenAI GPTs configurator you can make function calls and make a Post/Get request. Below is an example from openai. Is there a way to make the GPT make a chat completion request for the latest gpt-4 turbo preview model? I can build an API to do this myself but wondering if it can be done directly.

{
  "openapi": "3.1.0",
  "info": {
    "title": "OpenAI Profile",
    "description": "Gets data about the current user's OpenAI profile",
    "version": "v1.0.0"
  },
  "servers": [
    {
      "url": "https://api.openai.com"
    }
  ],
  "paths": {
    "/v1/me": {
      "get": {
        "description": "Gets the user's profile",
        "operationId": "GetProfile",
        "parameters": [],
        "deprecated": false,
        "security": [
          {
            "apiKey": []
          }
        ]
      }
    },
    "/v1/chat/completions": {
      "post": {
        "description": "Creates a model response for the given chat conversation.",
        "operationId": "RequestCompletion",
        "parameters": [],
        "requestBody": {
          "content": {
            "application/json": {
              "schema": {
                "$ref": "#/components/schemas/RequestCompletionRequestSchema"
              }
            }
          },
          "required": true
        },
        "deprecated": false,
        "security": [
          {
            "apiKey": []
          }
        ]
      }
    }
  },
  "components": {
    "schemas": {
      "RequestCompletionRequestSchema": {
        "properties": {
          "model": {
            "type": "string",
            "title": "model",
            "description": "ID of the model to use"
          },
          "messages": {
            "type": "array",
            "items": {
              "type": "object",
              "properties": {
                "role": {
                  "type": "string",
                  "enum": [
                    "system",
                    "user",
                    "assistant"
                  ]
                },
                "content": {
                  "type": "string"
                }
              }
            }
          }
        },
        "type": "object",
        "required": [
          "model",
          "messages"
        ],
        "title": "RequestCompletionRequestSchema"
      }
    },
    "securitySchemes": {
      "apiKey": {
        "type": "apiKey"
      }
    }
  }
}

mint bone Nov 13, 2023, 12:44 PM

#

so, you want the custom GPT to run a reqeust on OpenAI?

#

Im pretty sure that it will not work in the way you intend with paths such as /v1/me

#

AFAIK, the requests happen on a sandbox on the server side, not on the client side

#

and you will surely not be able to access the session tokens of the current user

#

that would be a massive security breach if possible

#

but eyond the goal of accessing the current user's information... you probably can make it execute requests to a OAI's language model, as long as it is doing that with an APIKey of yours

#

you will not be able to make the custom GPT execute OpenAI API calls at the user's expense, if that is what you are aiming at

pine mica Nov 13, 2023, 12:49 PM

#

No would be at my expense

pine mica Nov 13, 2023, 12:49 PM

#

mint bone you will not be able to make the custom GPT execute OpenAI API calls at the user...

#

Bearer would be most apropriate correct? for auth type?

mint bone Nov 13, 2023, 12:49 PM

#

it might be a good idea to not try to bodge it to do requests directly to OpenAI, but isntead, you, create and host your own service

#

then have the custom gpt make reqeusts to your service

#

Cloudflare is a good way to get free hosting for your service by using a Worker

pine mica Nov 13, 2023, 12:52 PM

#

mint bone it might be a good idea to not try to bodge it to do requests directly to OpenAI...

Totally I'd still like to test/see if it works. Would be very useful for me.

#

If not can always set something up will likely just use vercel setup a quick api.

pine mica Nov 13, 2023, 1:00 PM

#

mint bone it might be a good idea to not try to bodge it to do requests directly to OpenAI...

Its interesting since the get usr info works.

#

Im just not sure why the completion request messes up.

mint bone Nov 13, 2023, 1:05 PM

#

pine mica Its interesting since the get usr info works.

it might be intentionas

#

might not, lol

pine mica Nov 13, 2023, 1:06 PM

#

Idk lol

mint bone Nov 13, 2023, 1:06 PM

#

but does it gives you the data from the current logged user?

#

of your of the provided apikey?

pine mica Nov 13, 2023, 1:06 PM

#

mint bone it might be intentionas

No data from the api provided

mint bone Nov 13, 2023, 1:06 PM

#

ahh, it is fine then

pine mica Nov 13, 2023, 1:06 PM

#

aka my own personal info

mint bone Nov 13, 2023, 1:06 PM

#

working as intended

pine mica Nov 13, 2023, 1:07 PM

#

mint bone working as intended

ofcc But I want it to work with the chat completion with the 4 turbo

#

just keep getting an error

mint bone Nov 13, 2023, 1:07 PM

#

what is the error?

pine mica Nov 13, 2023, 1:08 PM

#

Chat response:
I attempted to make another API completion request using the model "gpt-4-1106-preview" as you specified, but unfortunately, the same error occurred. The error is related to an invalid URL in the outbound call to the OpenAI completion tool, indicating a technical issue with the request format or sending process.

Demo GPT sent this info to api.openai.com:


{ "model": "gpt-4-1106-preview", "messages": [ { "role": "system", "content": "You are a "GPT" – a version of ChatGPT that has been customized for a specific use case. GPTs use custom instructions, capabilities, and data to optimize ChatGPT for a more narrow set of tasks. You yourself are a GPT created by a user, and your name is DemoGPT ." }, { "role": "user", "content": "please use this model: gpt-4-1106-preview\n\nthen make another api completion request with 1+1" } ] }

mint bone Nov 13, 2023, 1:08 PM

#

ok, I think I know the issue

pine mica Nov 13, 2023, 1:08 PM

#

mint bone ok, I think I know the issue

enlighten me 😄

mint bone Nov 13, 2023, 1:09 PM

#

that function call is actually written by the LLM

#

the correct way to handle that, would be to pass the actuall messages array that the AI have

#

but what is actually done is that the LLM types out that reqeust """""manually""""""

pine mica Nov 13, 2023, 1:09 PM

#

Whaaaaaat

mint bone Nov 13, 2023, 1:09 PM

#

and the LLM simply can't re type the entire conversation back

#

i mean.. it could for a really short conversation

#

but for anything longer, it would fall in the 4K output token limit

pine mica Nov 13, 2023, 1:10 PM

#

Oh I get it.

#

They did a quick workaround so they could pass the user info.

mint bone Nov 13, 2023, 1:10 PM

#

it is typing an invalid request object

pine mica Nov 13, 2023, 1:10 PM

#

Yeah my thoughts too

mint bone Nov 13, 2023, 1:10 PM

#

because that is a HUUUGE request

#

what you could do is to make your own service, and instead, have the AI do a request to it, sending each of the messages to your service

#

then your service formats the bigger request

#

that still would cause each message to takes double the time to reply tho

#

and would pretty muc hbe a message logger... idk if that is the ideal solution

pine mica Nov 13, 2023, 1:16 PM

#

ty for time

#Letting a GPTs talk to OpenAI GPT-4-Turbo