Setting add
“update.enableWindowsBackgroundUpdates”: false
Setting add
“update.enableWindowsBackgroundUpdates”: false
Hello, Just wanted to share a snipped of a answer I got:
“I don’t see the actual content of index.ts in your message (it’s marked as “identical to original version” but the original isn’t shown)”
I think this is pretty insightful feedback from the llm itself.
I’m using Claude Sonet, and the chat is not very long. Two weeks ago this would have been very easy for Cursor to do, but now it can’t even get started.
@comonetso Does this work for you? I tried setting this and restarting Cursor twice/thrice, but it always still automatically updates to 0.43+ right after I close version 0.42.5 and reopen Cursor. Absolutely frustrating.
Same experience here. Upgrade and the magic cursor experience vanish.
Really frustrating. Is totally down for now. Hope you will fix that really soon.
it seems like new update came out, anyone tried to install it? was it worth it?
UPD: never mind, i’ve just received this update banner for no reason
I have to second this thread!
One new issue I have noticed is that sometimes when I add a context file to the chat, Claude doesnt see it. Even when directly asked about the file, it says it isnt seeing anything. So having to open the file I added to the chat history, copying it, and pasting it fixes it, but beats the purpose of using cursor instead of claude directly.
Hope this gets fixed, I was enjoying cursor, it was the best productivity accelerator, I kept recommending to programmers around me, but now its a source of frustration rather than help.
Same here…I tried to continue a project started at the weekend without the update. It was mindblowing…best Coding-Experience I ever had. But since the new update Cursor seems to be completely dumb. In a few steps my complete Project was destroyed…
The file structure messed up, Cursor seems to have no clue what this project should do (but it was mainly built by Cursor/Sonnet days before).
I don’t know what you guys made with your software, but I’ll hope you get that fixed soon.
So it went from “magic” to complete “fuckup” for no reason.
Bring back the old cursor.
I wondering if cursor was built with an LLM, probable… Also hanging for me and to add to that its overwriting code that it shouldn’t touch! This moring it has completely messed up my codebase. Yesterday it was overwriting code it was supposed to be documenting…
I give up but was about to pay for a subscription, can’t justify that with the problems I am seeing and reading about in this thread
I spent twice as many tokens redoing the texts that, after being sent to the chat, kept generating infinitely and having to cancel and re-ask the question to the chat.
It worked a few days ago, but not anymore. You can’t do anything crazy like modify the hosts file.
Yup, the last version was much more polished. The Chat isn’t providing horrible responses, but the composer is just straight up broken. It has a huge problem with duplicate code and does not understand project requirements. Even if you explicitly specify it not to do something for a specific reason it will continue on the way it wants. I’m pretty disappointed with this update, considering i am a paid user, my responses are just full of junk/ useless strategies which has pushed me to my usage limit, essentially from the responses breaking the code, then suggesting new response that break the code even further until you end up with a 500 line file for printing hello world.
For the most part pretty decent, but I’m sitting here asking myself why am i paying money for something that frequently breaks my code leaving me sweating hoping i pushed my last working code. idk man
This seems honestly like a pricing problem.
If we could we probably would all include the full codebase and run against Claude 3.5 sonnet (or pay for o1?)
The subscription only model limits Cursor, it caps their ability to expand with the project we are working on and I’m guessing here, but at the rate we’d all burn through compute with high token contexts it’s challenging for them to balance performance and context size with cost.
There are 2 options I can see here:
Some kind of either addon or credit system that could allow my context size to grow with my codebase (adding some kind of UX to show cost against some hard alerts/limits). Keeping the context small to conserve costs but allowing it to be elastic to my needs.
Include the file structure tree. This would help maybe contextualize the context lol. However it only shows the files, and thus what “might” be in them, vs what is actually in them. But still selecting key codebase elements to solve a prompt maybe more efficient.
When cursor was unbounded it was incredible, literally life changing - but with the complexity of me having to either correct the AI as it’s missing files, or remove duplicates or even re-add code in where suggestions remove needed code makes it less usable (but still powerful).
So wild! They must have changed a model on the backend overnight.
Today, I cannot use @Codebase
– it simply does not work. The “Generating…” window hangs indefinitely, no feedback.
Ooof, and just as I type that, the incident was just acknowledged two minutes ago: Cursor Status - Upstream Service Degradation - Codebase Search
Yep, it’s not something on our end unfortunately, OpenAI is not having a good day!
Thanks, Dan. How does OpenAI’s outage affect those of us experiencing issues with sonnet 3.5 selected?
Right now, we use OpenAI embeddings for codebase search.
This makes sense, thank you!
Probably minor but it would be nice to get a heads up about this in the Cursor IDE itself. I came here to complain, and had I not found this very post, I would be doing that.