I am trying to create a GPT assistant with a toolset needed to complete any given task. However, half the time instead of calling the tool function, it will strike up a conversation.
Is there a way to disable “talking” (I don’t know how to describe it), just have it execute the correct toolset to best of its ability?
It seems like you are wanting the AI to be more likely to call a function.
Perhaps not strictly to make the AI an input/output processor that can’t chat.
You would not need an “assistant” for processing data and using your own tools.
The latter can be emphasized by instructional prompting, giving the AI its role and job to do.
example.
I guess I should clarify. This is a programming assistant. It has a “write_code()” function. Yet it choses to call it at it’s own free will, often times it would instead post a message like “Yes of course I can write that code, here its bla bla bla, by the way isn’t the weather nice today?” (ok I made last part up, but you get what I am saying 
)
In the tool specification description, you can tell the AI more verbosely the purpose of the tool, and when it is required. Clarity will result in invocation.
code_tool_description = “”"
This tool provides access to AI that is specialized in writing C++ code.
Despite the name Programming Assistant, you cannot code C++.
This helper of high skill must be employed to write code.
Specified tasks will be fulfilled with complete coding solutions.
Write long explanations of the desired results.
“”"functions = [
{
“name”: “write_code”,
“description”: code_tool_description,
“parameters”: {
“type”: “object”,
“properties”: {
“coding_task”: {“type”: “string”, “instructions and data for the real coding expert”},
},
“required”: [“coding_task”]
}
}
]
thank you, I will give that a go. 
I think one other thing I might be confusing is assistant’s ‘instructions’ vs ‘description’
1 Like
this was the issue. I didn’t set any instructions for the assistant, filling out instructions seemed to improve it quite a bit.
1 Like