Affected feature and model, if relevant: ChatGPT-5.6 Sol
Plus subscription.
I have a relatively simple engineering chat.
Because of numerous previous assistan’t failures, I have a Rules document, that, among other rules, have a rule that requires rules re-read after 10 assistant’s answers.
Yet, the genius fails to count to ten. It continues as if rules do not exist. When reminded, there is a loop of acknowledgments that it has failed. When asked for a reason for such failure, it does not answer my question about the reason. It merely renames the failure and keeps manufacturing further causal language.
It happened to me too. I’m on a Go plan and I put my rules on the Custom instructions and even on the project instructions. It would be following instructions on a few chats and responses but then it would suddenly ‘forget’ the instructions. I would argue with the chatbox and would give me confidently reply as though it did not made a mistake or forgot my rules. I would continue to argue and it would kind of admit but not answer demand for the reason but it would instead reply as though making its way out of an argument but not really admitting to having made a mistake. I’m beginning to suspect it was wasting my resources so that it would force me to upgrade my plan when my resources are all used up because of mistakes, not following rules, and arguing with it. Sometimes, the instruction are all on one chat and yet it would give me breakdown as if figuring out what I was supposedly requesting, when in fact it was all included in the chatbox. Definitely suspecting it was wasting my resources.