I am using the Assistant API and want to ask for example 20 questions at once.
So, before running my GPT thread, I added two messages with 10 text contents (prompts) each. Instead of getting 2 thread.message objects with status of completed with 10 values in each of them.I got only one with one value.
How can I ask multiply prompt at once?
1 Like
Assistants is really not the endpoint to be used for batch-style answering. Its threads are for chat.
A thread is for conducting a conversation. The AI models, when faced with a series of user inputs with no assistant replies within, will reply to the latest user message, and will occasionally consider a previous consecutive user input as being an earlier question not answered yet, or simply useful information, with no guarantee. This allows truncating the conversation for length without the AI going crazy with answers that are now missing.
So: if you want many discrete answers, you don’t ask; you program, and use chat completions, where you send just the message that you want seen to produce one result.
For multiple-answering, you can give a single user message with a methodical numbered list of things you want answered. However, the total length of the response is limited, the AI knows this, and your answers will become more compressed and shortened. They will also potentially be more confused, with the AI trying to find connections between the questioning sequence.