What is going on with the new usage limit rollout? Just recently, a single prompt consumed my entire month’s worth of credits.
I had explicitly instructed the system not to re audit source code, files, or ZIP packages that had already been reviewed. I have multiple agents turned off, Sites disabled, and no other plugins or integrations active that would be causing unnecessary processing.
During that same run, one of the output downloads failed (Just a link with nothing attached, and nothing in sources or files library either) So I had to ask for the ZIP package Again. From what I understand, regenerating a failed download does not use additional tokens? (Don’t quote me on this, I haven’t received a degree in understanding GPT usage charts…Not sure anyone has, for that matter, but I digress.) I then had to purchase additional credits, three separate times at $10 each, and still ran out of usage. Mind you, all of this happened in just over 30 minutes. After that, I had to wait a full week for my credits to renew.
Yet, I still had to do a ton of debugging on account of numerous syntax errors, missing files, and other issues in the output.
Now there is an overall weekly limit along with five hour limits for each month. I can easily burn through the five hour limit very quickly without having any idea how much work I’ll actually get out of it; even when using lighter models.
I also find myself constantly having to tell GPT to get back on track, or stay on track, because it often thinks way too long on something, gets stuck, or returns messages saying “the system is thinking a bit longer than expected”.
After the five hour reset? When I return and prompt it to continue where it left off, it seems to consume an excessive amount of usage almost immediately because it has to catch back up with the previous context. However, I could be wrong In that assumption? Stranger things have happened.
The context within each chat needs to persist longer to avoid it relearning the same project every time usage resets. I think Open AI should try to get ahead of wasted processing and mayhaps make usage somewhat predictable, ultimately making these pesky limit bears somewhat manageable for development work.