There are many instances where we send requests to expensive models, and they either produce inaccurate results or make unwanted changes. In such cases, we often have to undo everything, but the user still bears the full cost. We should consider implementing a penalty for these situations. For example, if a user’s request costs 1x, and they need to undo it, only 0.5x should be charged. This approach would encourage AI model companies to improve their models and provide better responses.
it would also encourage exfiltrating the answers and then undoing all the work to get free stuff
It still charges half. This applies only to agents, not to chats.
still not gonna happen
Are you a member of an AI advocacy organization? 
Whenever I get a bad response, I try to create a memory rule that prevents it in the future.
It seems to reduce the errors quite a lot.
there just not going to do it, no matter the upvotes
Cursor’s position will be “just use auto”
Great idea! Even just a counter for users would be nice so we could see how happy we are with each model.