I’m implementing a multi-agent system and I’ve noticed that the agent is too proactive do I have explicitly ask it many times, sometimes even as the last sentence of the chat, to DO MOT START implementing anything until I say so explicitly.
Still, it continues saying “you’re right. I should stop” and then it decided to continue implementing some tasks it had decided that needs to be implemented.
This behavior is unacceptable because it breaks my design and then I need to undo what it does. And also because it’s consuming tokens on unapproved tasks.
How can I stop it from happening and how can I get my money back?
As you can see, it is aware that it’s not following my instructions. Only option is to roll back but, what about the consumption (close to 4 credits)???
Thanks for sharing. The problem is that it is ignoring my commands so it doesn’t matter if I set surviving rules or invoice one set of specific commands to all my prompts. It will continue doing what it wants.
Honestly, I’m testing different tools for my product and this one unfortunately has failed the test. I can’t afford using a tool that mess with the design and spend the money doing so.
on my macbook i use textreplacement from keyoard and and for thinking questions this: “Do not change anything, do not generate, and list. Focus only on the facts and leave out assumptions. List all open points explicitly if the solution is incomplete. And do not forget: there is no users but only public.user, and we have a schema that you use if you need it. And really do not change and really do not generate and only answer and only explain – this is important!”
