Have any of you already tried a “recursive” style function call mechanism?
I’m not talking about several calls at once.
Let me explain: in a way, it’s the result of a function call that directly leads to another function call in a context that expands and depends on each other.
I’m getting pretty good results with ChatGPT3.5 Turbo (preview), but with ChatGPT4 Turbo (preview), it’s causing a timeout every time.
We don’t need to “try” it, because that’s what the AI already does if thinks multiple calls are useful, such as try two different search engines, try again if an error was encountered, etc.
The non-responsive gpt-4-preview is pretty much spot-on. You can set a timeout on not getting a response back, and then try again.
client = OpenAI(timeout=30) # use the longest wait your normal responses may require.
I was testing the other day how to make an agent. If the user makes a complex query, it will spawn several tasks and each task can spawn subtasks. After several tests, I immediately reached my TPM limit lol.