Hi,
I tried to use the gpt-3.5-turbo-1106 on the API and in the playground and noticed that it calls functions much more agressivelly than gpt-3.5-turbo or gpt-3.5-turbo-16k.
This is not a welcome feature, because it calls functions where it should not call. This is only with 3.5, the new gpt-4 (gpt-4-1106-preview) do not do this.
Anyone else noticed this behaviour?
Make sure you use the required property to indicate the required params for the functions. Also, try adding some language like, âOnly call a function when all required parameters are fulfilledâ.
Did try that, I put some examples for the expected params and the alert to not call the function if they are not present, but on gpt-3.5-turbo-1106 this is less efective than on other 3.5 models
same problem here , i have a function calling for generating images, itâs always call this function in almost prompt, like:
âby Laurence Sterne, was a pioneering work in the invention of the stream-of-consciousness novel, which became an influential novelistic genre in the hands of Virginia Woolf and James Joyce.â
I am experiencing same problem too. I have list of about 10 functions. For all the user prompts, it just calls the most relevant function even though function call is not required. Iâve tested comparatively with gpt-3.5-turbo-0613 too. 0613 version works perfectly fine for the same prompt and same tools schema.
I think the new model heavily relies/biased on function call. Hope they will fix it soon.