EDIT: I tried building the GPT using the interaction with chatgpt instead of using the "Instructions" on the back end. I wonder if that is causing these problems as it is continously updating the instructions. I think this is where the problem is that it struggles to keep all previous instructions completely isolated and update only those that need to be. I will try rebuilding using only the text based instructions part of the configuration section ================================== Have been building a custom GPT and am finding (what I belive are some bugs/issues - yes I know it is in beta) 1. I configure the "conversation starters" to exactly how I want them to be and without warning or reason the builder just updates them and changes them even after I explicitly instructed it not to 2. Previously saved instructions are destroyed/changed wrongly after an update even though I was not setting instructions for that prompt 3. There are various little annoyances/bugs like the number of prompts displayed are different to the number that have been configured From my experience I think this is only good for a "one prompt" tool like the one in Jason West video about the "Auction" thing wher eit only does one simple thing in response to the users one action. Anything multiple prompts and each prompt has a different set of instructions seems not to work too well...this is my experience with it anyway For reference - I built a GPT to help me with some document management. There are 5 potential tasks a users wants to perform. Each tasks has a series of questions the tool must ask to establish what to do. This is then followed by the user uploading the content and the tool then either rewording, reformatting or doing a variety of tasks to change the content to the users specification. There are at most 3-4 questions. The actual output and task is well done but if I update the tool with some further instructions it will randomly then just change all the prompts and lose a whole lot of the previously understood instructions. It does not seem to have an "rollback"/versioning facility where it understand the previous version and can roll back to that. At this time seems a bit of a waste of time for anything but extremely basic simple tasks

#Copywriting I click on the link and the following screen

#Blog Posts excellent straight forward tutorial. Still a bit blown away that this is possible

I have 3 different articles (urls) that talk about the same thing but are all slightly different with slightly different content but the main content about 70% of each article talks about the same thing. I would like to give the urls to chatGPT and have it "consolidate" the main points from all 3 articles. Is there some trick or prompt to make this happen or atleast shortcut it somehow?