Hello,
I’ve been using the PRO version of Cursor for a while now, and I’m still disappointed with the AI’s behavior.
It tends to forget things or go in circles.
For example, Cursor makes me test a curl on port 5050, and from time to time, it switches back to port 3000 for no reason.
Another example: Cursor tells me to run commands and then forgets I already ran them just a few prompts later.
One last thing: sometimes I give it the database values directly to make things easier, and it still generates commands to go check them.
That said, it’s still a major breakthrough — and seeing what version 1.2 can already do makes you excited for version 3 or 4.
But these small annoyances are a real pain and an ocean of frustration!
Vince
Hey Vince.
If you’re encountering unexpected behaviour from chat outputs, I’d recommend:
- Reloading the chat often cmd+N (mac) / ctrl+N (windows)
- Splitting tasks into smaller chunks.
- Supplying only the necessary files.
- Make sure the LLM you’re using is best suited to your use case
These tend to solve most issues that can appear from LLM hallucinations and overly large context windows.
You can customise a fair bit also with user rules which can be super helpful