I am using GPT's API to feed it a series of generated yes-or-no questions (which is being fed into a Boolean type data). When I ask GPT the question without any instructions, it gives a logical answer. However, when I give instructions to say either yes or no, it gives a response which both a) doesn't make logical sense and b) contradicts its original answer. Observe the Python code below: from openai import OpenAI client = OpenAI( api_key=<api key> ) model="gpt-4o-mini" instruct="give response as either 'yes' or 'no'" query = "is 'Incredibles 2' (2018) either a remake or set in the same continuity as any movie made by Walt Disney Pictures prior to 2018?" print(query) response = client.responses.create(model=model, instructions="", input=query) print(response.output_text) response = client.responses.create(model=model, instructions=instruct, input=query) print(response.output_text) The first response, with no instructions, says: "Incredibles 2" (2018) is not a remake; rather, it is a direct sequel to the original "The Incredibles," which was released in 2004. Both films are set in the same continuity and universe that was created by Pixar Animation Studios, which is a subsidiary of Walt Disney Pictures. The sequel continues the story of the Parr family and further explores themes of heroism and family dynamics. Thus, the answer should be "yes" (because it is set in the same continuity as Incredibles 1). But the second response, specifying the answer should be yes-or-no, instead says: No. Which makes no sense, and contradicts the first response. What should I do in order to get the kind of output that I expect? I suppose I could have said "is it either a remake or a sequel?" but I also want to get a positive response for movies that are not a direct sequel, but set in the same extended universe (like the MCU). Continue reading...
