AFIK, gpt-4-32k is not currently supported by OpenAI's API. I'd love to have it but alas... My custom API calling code has the body of the HTTPRepsonse from a URLRead, the following:

{
    \"error\": {
        \"message\": \"The model: `gpt-4-32k` does not exist\",
        \"type\": \"invalid_request_error\",
        \"param\": null,
        \"code\": \"model_not_found\"
    }
}

Hi Russ, long time since we worked on our Wolfram Solutions project together -- I still have all the code ;).

In the Wolfram solutions group, we have been working on quite a number of different demonstrations of how to use LLM's's in a variety of ways. One offshoot of that that I put out publicly on the paclet repository is this:

ChatGPTPromptExplorer

And in that paclet all the code is available -- including a custom GPT API calling function, so one can poke around in the code to see how I managed to do things,

And an offshoot of the various custom functions for directly creating chatbooks is this little bit of code that I pulled out of one of my other demonstration projects:

createAChatbook[persona_String,model_String]:=
    CreateDocument[
       {},

       StyleDefinitions -> "Chatbook.nb",
       DefaultNewCellStyle->"ChatInput",
       TaggingRules ->
         <|
           "ChatNotebookSettings" -> 
               <|
                  "Model" -> model,
                  "LLMEvaluator" -> persona,
                  "Assistance" -> True,
                  "ChatDrivenNotebook" -> True
               |>
          |>
    ]

If you execute it with, say,

createAChatbook["PlainChat", "gpt-3.5-turbo-16k"]

or this

createAChatbook["PlainChat", "gpt-4"]

it should work (it does for me ;-) as I have a gpt-4 enabled api key ).

If you try

createAChatbook["PlainChat", "gpt-4-32k"]

OpenAI will yell back at you.

Anyway, that bit of code may be helpful to folks when trying to programmatically create a chatbook. I actually don't know what the official code is within the chat book paclet that generates ChatBooks from the menu, so the approach in a bit of code that I have above may or may not processed overtime. But it's fun to play with.

And some of the options that I have in there may or may not work correctly (e.g. "ChatDrivenNotebook") but you can experiment...

@David Reiss

Hi David! by chance can you tell me the secret per my note above?

ya. i mean we already use Palm/bard with the LLMfunctions and PalmLink paclet in the paclet repository.

but there is a way to set the GPT based models... i just don't know the location that the front end is pulling it from.