Closed HITLittleZheng closed 5 months ago
Yes, why not. In our experiments, when compared against ChatGPT4, ChatGPT3.5 is slightly less accurate as a judge so I would suggest manually verifying a few annotations before running it for large-scale experiments.
Dear author, in performing the second step, I don't have the API for gpt4, can I use the API for chatgpt instead?