Everything I do, results come back with the wrong language and data on projects that have nothing to do with anything I have ever worked on. At this point it’s around 80% of the requests – it’s un-usable at this point. I have seen this a little (maybe twice in a few years) but now it’s all the time.
Steps to Reproduce
Easy, I just do a request.
Expected Behavior
That my requests are logical. They even have different languages.
Hi @Carlvon1800 Thank you for the post and for specifying the models where you’ve seen this. This is a known model behavior issue that we are aware of and working to reduce. Trying another model is a good workaround, tweaking the prompt, and making sure that the context window is not too filled up are all good steps to take. We’ve also seen behavior like this after repeated self-summarizations.
Thanks Kevin, I have switched to Composer 2.5 Fast for now and it seems to be working. It may be something with the AUTO system (I assume other vendors models).
Thank you for following up! If you have a particular request that seems bad you want us to take a look at feel free to share the request ID (ideally with privacy mode disabled before you make the request). Here’s how to do that:
To disable Privacy Mode:
Open Cursor Settings with Cmd+Shift+J on macOS or Ctrl+Shift+J on Windows/Linux.
Go to General.
Turn Privacy Mode off / switch to Share Data.
To get the Request ID:
Open the relevant conversation in the Chat sidebar.
