I was doing some work today, building a glossary, and saw this new User Interface (UI) for preferred responses. It was responding at the same time in two different boxes. It then allowed me to click on the prefered output. I assume they are using this feedback to improve the responses of ChatGPT-4.
Unfortunately, the output on the left failed and the output on the right was trying to complete the task with python, but it was showing in plain text. We have to use GPT-4 because of all the integrated bells and whistles, but I really do think GPT-3.5 has much better outputs with fewer errors and rework required.