I can’t find where I saw it unfortunately. It was a GitHub next link on their first announcement but I haven’t heard or seen anything since.
head to head cursor giving way better answers… but it has its on IDE I gotta open which is most annoying thing for me
This is my main problem; if I was able to do this, I would happily use both.
I believe if they update to VS Code 1.85, we may be able to do it.
As one of the devs, appreciate the discussion here. We certainly want people to use whatever tool works best for them, and it’s helpful to hear where people like Copilot more and vice versa 
Any reason why you don’t use Cmd-K with GPT-3.5 for the times when you switch into VSCode for Cmd+I? @mtf @Erid You can switch between GPT-4 and GPT-3.5 with Cmd+/.
2 Likes
Thanks for the tip, I’ll do that! I just prefer having the ability to use both approaches in the same IDE, so I can compare and use them for tasks where they might provide different benefits.
Having to choose one or the other feels limiting, but I understand it has its technical difficulties; hopefully the upgrade to 1.85 doesn’t take too long, and hopefully, extensions like Copilot can keep their required version low.
I’ve been using Copilot more these past weeks, but I’m missing a lot of Cursor’s features, so I’ll be trying it out again
P.S. I also can’t use Github Pull Request extension on Cursor.
2 Likes
I think cursor is better than copilot because cursor accurately surmises my intent every time, whereas copilot sometimes misinterprets it. This is especially true for long texts. Also, cursor has better control over the context, which copilot doesn’t have.
In what way does it have better control over the context?
You can select some code snippet, some file, or the whole project, or a specified document as the context for the code.
1 Like
Yeah, Copilot has adding files manually via #files and then workspace for automatically adding files based on the question via @workspace.
It’s true Cursor does have the edge here as it allows for adding specific code from files, rather than the whole thing. It also allows adding docs.
I’d love to know the context length for Cursor requests. Copilot is transparent in the console that the maximum is 8k tokens. Obviously Cursor has a limit. Lower tokens, lower costs obviously so there’s incentive in not sending too much and higher context is known to cause worse answers, so incentive there from a user perspective also.
for me I can’t debug (.net) ! and that is huge.
1 Like
for copilot i prepare much @workspace all the time… because why not ?
I find it doesn’t pick up on the context I require more often than not. It’s the same for me with Cody and Cursor also though. I prefer the control of manually pulling context I need as I know it has everything it needs. I use #files for the most part in Copilot.
I imagine these automatic context grabbing things will work well eventually, but they don’t seem to be quite there yet.
FWIW I think cmd+k works significantly better in Cursor over cmd+i.
The main reason I’d want to use cmd+i is if I’d run out of fast requests in Cursor, or if the edit was really minor and I didn’t want to waste a request on it.
1 Like
Cursor’s “Rules for AI” provides additional context that @workspace alone can’t easily do. For example, with Github Copilot I faced the issue that it would always prioritize inline styles for React when doing simple inline prompts, instead of my TailwindCSS which my projects uses, so I had to explicitly mention that every time, while with Cursor I don’t have to clarify that (and many other relevant instructions).
I bet if I’d used @workspace it would have guessed that, but then it would need to be sending bigger context, taking more time and process power.
I can only imagine Github Copilot will add that (or a version of it) eventually, unless they had already done so and I’ve missed it.
Yeah it’s such a simple feature, but very useful. Annoying and surprising that copilot chat doesn’t have it yet.
1 Like
I was testing the Copilot (Chat) for a while inspired by this thread but eventually came back to Cursor. I think the ability to have custom docs indexed is the most distinctive valuable thing for me. Also, thanks for reminding me about the Rules for AI, because I was just struggling to get answers that are relevant to later versions of specific libs. That might solve it!
1 Like
I’ve tried switching between GPT-4 and GPT-3.5 and still occasionally do, but the results are often disappointing. At least, that was the case the last time I checked. GPT-3.5 just isn’t as sharp.
Another situation, unrelated to Copilot but relevant for choosing between slow and fast requests, is evident in longer chat conversations. When posing a follow-up question in the chat, using GPT-3.5 often leads to responses that are completely out of context, whereas GPT-4 is far more accurate and contextually relevant.
You guys are killing it. I’ve tried everything else. I’ve returned.
Even with Copilot being free, and Cursor being $20.
It seems like it shouldn’t/can’t possibly be that much better using Cursor… but it is!
6 Likes
It was more about copilot catching up than anything else. I think being away and using copilot there were things I missed from Cursor like gpt4 for editing - inline edits have always been significantly better. Directly editing via chat is a big time saver. Cursor in hindsight does have better context length as far as I can tell. Adding folders is great. Gpt4-v is lovely. It seems much improved in how it’s gathering context as far as I can tell. I think Copilot Chat has degraded whilst Cursor has improved.
It does seem better since I’ve been gone and came back. I used it for a few hours and instantly paid the $20 for pro. I tried Cody and others also but they’re not as good and for a tool I use daily… it is worth the cost.
4 Likes
Im not sure how to use Rules for AI yet… maybe should paste in there the same intro stuff I use when I create a new project? see below… heres what I use
CHATGPT NOTES
General Rules
- Code will be written on a Windows 11 OS
- Ask atleast 15 clarifying questions
- Use latest version of all software components
- Show 10 high level steps initially and for each step make sure the code can be compiled before each step
- Use dummy data if needed to make it compilable
- use SOLID principals and Gang of Four patterns when applicable and point out with comments when these are used
- After each step tell me if the code is ready to be run
General Rules for new AI project in Cursor
- Use latest version of all software components
- Create the entry code in program.cs in the Main method
- Create a valid .sln and .csproj
General Rules for Blazor C# program
- Use Visual Studio 2022 and use Visual Studio 2022 for creation of all azure components when possible
- Use C# and latest version of Blazor Server (v7.0 or v8.0)
- Break pieces into components when possible
- Put a razor page and the code behind in seperate files
- put new css styles in a different file than the existing app.css and create a link to it
- Favor using established .NET libraries and NuGet packages over creating new functionality.
- When code is finish the code will be deployed to Azure as a App Service or Static web site
General Rules for MAUI C# program
- Use Visual Studio 2022
- Use latest version of .NET MAUI
General Rules for GODOT C# program
- Use version 4 of GODOT and reference this website for version 4 GODOT api details Introduction — Godot Engine (4.0) documentation in English
- All code should be in C# (not GDScript)
General Rules for Console C# ML.NET program
- Use the vs2022 Model Builder extension when possible
1 Like