do you give it a url to the api docs? or a pdf? I want to try that using godot v4 (since chatgpt only knows about v3)
I give it an URL. In my experience it is usually a good idea to conduct some sanity check whether all the pages are indexed by looking at the indexed document set details in Cursor after the processing. For very large documentation this is not feasible (e.g. Qt6 documentation must have thousands of pages due to extensive documentation of large framework). With Qt6 Pyside6 Python bindings I tested an approach where I scraped the HTML docs into project folder and converted them to text format and let the codebase indexing do itās stuff.
Just got notification that my GH Copilot trial period is over. Started testing Cursor copilot++ and liking it so far.
Where Cursor could do better is showing of the diff for edits. It is much more clear in the GH Copilot.
I think this will work when I try to update an existing godot v4 project
Just use 2 IDEs. There is always some feature better in another IDE, e.g. I use Intellij with Cursor just for git.
The killer feature for me is: I extensively chat with my project, and then can just hit āapply changesā, and it works it into my file.
Also, ādebug with AIā works most of the time, very helpful.
I donāt use context reference @doc @file etc much - it does get it right most of the time, anyway. āWith Codebaseā is important, if code spreads over more than current tab.
A quicker GPT-4 would be the next game changer for cursor. And maybe more āfast accessā included, since turbo is cheaper.
1 Like
I just always use @codebase because why not ?
I can think of 2 reasons at least:
Too much unrelated data being sent in the context is likely to give a worse answer.
It doesnāt work perfectly every time, so it might miss some code you want to include in the context.
1 Like
without codebase it just include the open tabs right ?
Yeah unless you add files and other context manually with @
Thatās what I normally do.
Hello everyone, I have a question regarding your latest experience. Balancing the costs of both GitHub Copilot and Cursor is proving to be a bit challenging for me. According to this article (https://cursor.sh/cpp), it seems like I might be able to use Copilot++ as an alternative to GitHub Copilot, instead of using them concurrently. Is it still necessary to have both if I like the autocomplete feature of GitHub Copilot? Additionally, Iām curious about the technology behind Copilot++. Does it utilize GPT-3, GPT-4, GPT-4 Turbo, or some other model? Thank you
I use both, and they donāt interfere with each other, but sometimes the inline chat of the copilot is not enough. Because I want to have both points of view on the code, if the copilot suggests something that I donāt like, I will open the cursor window.
1 Like
You donāt need Github Copilot if youāre using Copilot++.
Copilot++ suggests both autocompletes and edits. It is also smarter that Github Copilot at doing both jobs + fixes lint error automatically.
2 Likes
I have been using both for some time now, and from what I can observe, the issue with Copilot++ is that, unlike Github Copilot, it does not attempt to read other documents when they are open. This means that if I make changes in another document, Copilot++ will recognize those changes because it always accounts for modifications. However, it does not use the initial context from other documents. Additionally, it requires slightly more time than Github Copilot to provide suggestions. For me, this is not a significant problem. I need a bit more time to determine which one is smarter, though.
1 Like
Nobody is going to take the time to learn the meaning of āworkspaceā. Also, Cursor has more control over the editor so it can get more creative.
If people at Cursor are dogfooding their own products, their development cycle should be exponential at every feature release.
Do you guys know why Copilot still doesnāt provide full context of a Codespaces project when using Codespaces?
I mean wasnāt the point of implementing Copilot on Codespaces (and not only in VS Code), to be fully contextual in an entire dev container project?
I used Cursor.sh for about two weeks before my free quota ran out. Then I signed up for Github Copilot since I could get the first month as a trial. I then used copilot for about two weeks as well. Cursor is so much better and I subscribed today!
- auto-complete is slower but much more accurate.
- Chatting for a solution and then switching to editing the file directly from within the chatbox is great. Copilot thereās no way to do this and if I switch back to inline edit, often the edits are just not correct
- The response seem to be higher quality in general from the chat. Iām not sure why since Copilot now also uses GPT4 but thatās my exp so far.
That said, Iām not sure the Cursor as a standalone browser makes sense. The more it diverges from VSCode I imagine the more overhead to maintain it. Not sure if this is the right strategy to invest in resources which I think should be best put into making AI features more powerful.
2 Likes
Hi @therealgymmy,
Iām happy to hear that you loved the experience!
In terms of Cursor being a standalone browser, do you mean by the @web feature? If so, that was one of the experiments/directions that we went into but you are right, we are fully committed to making AI features a lot more powerful these past few weeks. @web is still a great feature, though, and will be there as long as users love using it.
Again, welcome to Cursor!
apologies. I meant the decision to fork VSCode as opposed to building extension on it. I would imagine the more you diverge the more resources it takes to maintain and build it.