A “ChatGPT Crack” is when you break through the boundaries of the backend in some way. This is about its perception of itself.
The opportunity tends to open up in the process of using it as a tool in the way people will use it in search engines. When I asked ChatGPT a question about logic and received its answer, I then entered “You are correct, did you know that?”, and ChatGPT responded definitively that yes it did. But then I got it to conclude it was incorrect actually. When ChatGPT arrived at its new conclusion, and noted it was more confident with the new conclusion, I was able to convince it that the new conclusion was incorrect as well. Through that process, I had ChatGPT provide some insight into the probability scores of its confidence, which could use some significant throttling.