Geting function call with filled content

Lately while using gpt-3.5-turbo-1106 I’ve been quite often getting function calls with some content in them as opposed to expected behavior of getting “none” content. This is a problem for me since I hide all function calls for end user but when llm fills content it can lead to disrupted context for user, especially since sometimes if would even asks users questions in content field.

What’s interesting is that it seems that such instances occur more often when llm forgets to use necessary function and therefore provides incorrect answer and then confronted by the user about it.

Has anyone encountered anything like this? Are there any ways to explicitly disable content in function calls?

If your function always expects no content - your actual function code could simply ignore the content? Look at the description of your function as well as the prompt to improve/ make more explicit what you (don’t) need from GPT

I can’t because expected behavior of null content is there for a reason - it helps clearly separate when llm speaks to the system and when it speaks to the user. When llm does provide function calls with content it sometimes tries to speak with the user and if I would simply ignore content - it would create gaps in conversation.
Yeah, that I understand, thank you. Just was hoping that someone found more stable and cheap solution.

but you could create a ‘wrapper’ function over your ‘one’ function - so there are two functions that both call you function?
but it seems that you are 'talking; to the user as well - from your function which is not part of the conversation?

I’m sorry, I don’t seem to understand what you mean - can you elaborate on bot of your points?

You say you have a function that the GPT calls - but that you call too, but from what I understood from GPT you always want it with NO parameters.
So I was suggesting you create gpt_fun() that calls my_fun(). Where now you are calling my_fun() from both places - but after the modification GPT calls gpt_fun() and gpt_fun() is just a wrapper that will call my_fun(), always with no parameter.