I use Web ChatGPT with my Pro subscription to ideate and design, and both GPT Desktop app and Codex CLI on Windows to do work on a local machine. Part of my workflow is to have my ideation/design context transition to adding in orchestration responsibility until I can do a full hand off to a local context.
Well, that workflow is turning out to have recently broken by a change in behavior.
Long outputs from Web ChatGPT, at least into copyable blocks, are getting truncated. So when I have the Web ChatGPT create a prompt for a local context, the prompt is truncated. Even when I had it split a prompt in 2, the 2nd part was truncated. Sometimes I get a button to generate the rest of the prompt, but usually it just gets cutoff mid-line. I can no longer trust that Web ChatGPT has created a prompt I can simply paste into my local context.
I even tried having it split the prompt into sections of 10K or fewer words. Did not solve the problem. Part 2 of 3 was arbitrarily truncated and part 3 never showed up.
I hope that there is an understanding that Pro subscribers have, in general, Professional concerns and need for reliability. I can also understand OpenAI’s needs for conservation and frugality. If this is really necessary to keep costs down, then give us a mechanism to preserve our workflows and give us all the information we need to adapt before it is rolled out. I have spent 10’s of thousands of tokens trying to work around this and am feeling blindsided and frustrated.
Also, I let the built-in AI pick the metadata for this post…it seems lackluster.
Thank you for your time and trouble in dealing with issues generated by policies you ( overwhelmingly likely ) have little to no say in.