Hi, I have the following scenario in a chatbot I’m building with GPT4o model:
Query 1: what (A) is responsible for (B)? This means (B) could be caused be a bunch of things, and (A) is the primary reason. This is working.
Query 2: is it (C)? Here C is a particular attribute of B. In using GPT client’s chat completion APIs, I noticed here “it” is always inferred as (A), so the answer is always off. However, testing it in chatgpt website, “it” is always correctly identified as (B). I’m curious how chatgpt realized this, and I’d like to build that into my own chat expereince. Thanks!