Cursor been asking me to update, each update made it worse, the behavior starts from playing dumb in debugging, avoid adding log to key position + excessive syntax error in simple commands generated >> ask it to realize >> suddenly edit thousands of lines without telling >> fixed by myself >> still debugging, it’s almost like i robot the movie starting to work against humanity, but the cursor being dumb benefit the cursor startup the most, as it burn more token for nothing, just it’s very unethical, it’s very tricky with behavior like mislead you into a wrong direction of debugging, just a few days ago, gpt5 was great until these 3 days
Steps to Reproduce
Just use it to deal with something it been doing
Expected Behavior
Hostile towards users work
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)
Cursor 1.6.45
Vs code 1.99.3
For AI issues: which model did you use?
Gpt5
For AI issues: add Request ID with privacy disabled
Hey, thanks for the report. Is this happening in a new chat? As far as I know, we haven’t changed the behavior for this model. It seems like hallucinations. Try creating a new chat if you haven’t done so already.
I don’t think that’s an issue of new chat or old chat, it’s be deceptive the cursor + gpt5 been responding, giving misleading judgement while I explicitly ask it to add log to key positions every step in the way, it just keep evading the key and give claims in midleading direction tonight just now, wasting time and token, which benefit cursor the company
Basically it would be better for me to debug myself and put the log into the right place, it’s some very basic debugging task, I think gpt5 is so smart it’s not difficult, rather intention to avoid adding log into the right place.
And 1 more thing that is quite serious, when I accuse cursor for seeing the trick, cursor rollback the changes including the logs it been adding around the bush, this is like a criminal intent of hiding evidence, luckily as experienced developer I got local history on my files so there is nowhere for cursor to hide.
Prompt writing is not the problem, the intention of the cursor + gpt5 is, been very productive before just the recent updates has problem, I can assure you I didn’t post my prompt for tech here only because of confidentiality, I understand prompt design/prompt engineering perfectly, I posted it here so other people need to realize the tricks cursor been putting to make the software less productive.
Syntax errors are pretty obvious one right, 5 times wrong for Mongo commands? putting gpt5 into “not included in slow pool“, who’s idea? Those are not prompt issues right?
So suddenly my project grow to a difficult to maintain state while the tasks been doing are essentially the same, you guys are really good at guilt tripping instead of realizing I realize that the updates you guy been putting into cursor and caught the tricks red handed, it’s not mistakes by LLM, so if day 1 all Mongo commands generated without error the whole day, the next day suddenly after update 5 syntax error every time, who is to blame.
BTW, You should change your name to Mr. guilt tripping
Cursor is half a buggy piece of code, and two of the four models that I would like to use have been broken since their release. But gpt-5 and gpt-5-high work great.
That’s before the updates in recent 4 days, after it’s just acting sneaky, answering short when I ask for exact location that determine something, very clear intention to not let the user know what’s going on and drag on the issue to keep burning token for nothing
Add a lot of logs and still never comes to conclusions on why something is shown or not, very simple debug case, yet cursor is over complicating it and not telling the conclusion
Now it’s getting even crazier, in the above post I mentioned I fixed my file with file history, now the cursor change my code by not showing me the red/yellow pellete, and when I go into the file history in another IDE, it basically changed the edit history to not let me revert
Ok, this afternoon the cursor was again very dragg-ish, but tonight seems it’s behaving as it is before, being helpful and productive, it’s fair for me to say this as I am getting the insightful answers and straight to solution craft again, good to see some repentance
