The funny thing is, if you point out its mistakes, it often does better on subsequent attempts.

Or it get stuck in an endless loop of two different but wrong solutions.

Me: This is my system, version x. I want to achieve this.

ChatGpt: Here’s the solution.

Me: But this only works with Version y of given system, not x

ChatGpt: <Apology> Try this.

Me: This is using a method that never existed in the framework.

ChatGpt: <Apology> <Gives first solution again>