We would like to see a Hybrid model Copilot where it can learn our local code with local AI engine and communicate the server for additional optimizations, that would be an ideal AI assistant for developers
@@shirosurfer8864 invest in your machine. I develop always on a high end machine. If you put a low end nvidia card into your pc, it doesn't cost that much. Currently I have a ryzen 7950, 128 GB RAM (DDR6) with an NVIDIA Geforce 3060. If you upgrade your cpu every year you save a lot of money as lon as you can use the mainboard.
My biggest beef with CoPilot chat is that I often need more time to explain myself more to the AI than to do it myself. If I have 2 similar classes but with like 20 properties I would ask copilot to write a mapper between these classes. The response it gives me is how to abstractly set up an extension method.. as in public static MyEntity ToEntity(the MyDto dp){ return new MyEntity(){ Id =dto.ID // rest of conversion } Like for fuck sake. The rest of the conversion is what i want him to do. And then when i tell him i want to map all the possible properties as best as it can then it will still fuck up somehow. Yet when I copy paste both classes into chat gtp I get the response i expect. I have paid for Copilot since the start but damn chat never fails to annoy me.
It's because the ram it takes to answer do these deterministic contexts it's huge huuuuuge I'd imagine like 500gb ram for a small project to get a right answer I think they really just use one class context and then a few or two more at max for each answer
@@shirosurfer8864 fair enough, im just saying what makes me dislike copilot chat. On the other had it could be multi transactional as in. I request a mappint between ClassX and ClassY. Then it will request both those files and VS finds them and sends them over
QQ - What version of .Net are you using in the demo? My GitHub CoPilot Chat response is saying: "As of the time of this response, .NET 7 has not been released yet. The latest stable version of .NET is .NET 6.0.".
@@marek19922209 and how do you do that? When you do an ATL+/ it brings up a box that you type in a question. No close button (Escape only), you can't resize it, you can't move it. Copilot Chat is in it's own window, but not the Ask Copilot window. It's like the built the prototype for it and someone said "Ship it!". The rest of Visual Studio is very polished and great. Copilot, not so much.
We would like to see a Hybrid model Copilot where it can learn our local code with local AI engine and communicate the server for additional optimizations, that would be an ideal AI assistant for developers
Yes I agree but I can only imagine the terrorífic amount of ram it'd take to take that much data into account for just sending one context scary
@@shirosurfer8864 invest in your machine. I develop always on a high end machine. If you put a low end nvidia card into your pc, it doesn't cost that much. Currently I have a ryzen 7950, 128 GB RAM (DDR6) with an NVIDIA Geforce 3060. If you upgrade your cpu every year you save a lot of money as lon as you can use the mainboard.
@@shirosurfer8864 Does not require that much bro, Tabnine is already doing it, only thing is their language models are good enough compared to CodexAI
thanks man! appreciate that beautiful explaination
Thank you! I loved the phrase about the Copilot output: “Do not trust it until it is tested”. 😂
Nice demo thanks!
very nice
Sir i couldnt open coplot propting box. I tryed Alt+k, alt+/ but i couldnt open it. How did u open it?
My biggest beef with CoPilot chat is that I often need more time to explain myself more to the AI than to do it myself. If I have 2 similar classes but with like 20 properties I would ask copilot to write a mapper between these classes. The response it gives me is how to abstractly set up an extension method.. as in
public static MyEntity ToEntity(the MyDto dp){
return new MyEntity(){
Id =dto.ID
// rest of conversion
}
Like for fuck sake. The rest of the conversion is what i want him to do. And then when i tell him i want to map all the possible properties as best as it can then it will still fuck up somehow. Yet when I copy paste both classes into chat gtp I get the response i expect.
I have paid for Copilot since the start but damn chat never fails to annoy me.
It's because the ram it takes to answer do these deterministic contexts it's huge huuuuuge I'd imagine like 500gb ram for a small project to get a right answer
I think they really just use one class context and then a few or two more at max for each answer
@@shirosurfer8864 fair enough, im just saying what makes me dislike copilot chat.
On the other had it could be multi transactional as in.
I request a mappint between ClassX and ClassY. Then it will request both those files and VS finds them and sends them over
QQ - What version of .Net are you using in the demo? My GitHub CoPilot Chat response is saying: "As of the time of this response, .NET 7 has not been released yet. The latest stable version of .NET is .NET 6.0.".
Such a handsome explaination uwu ty very much kiss kiss hehehe c:
Programming vs Promting and Check if Ai writes it correctly
what about rider?
That UI is terrible, please fix it. It should be in it's own window.
you can move it to its own window if you want actually
@@marek19922209 and how do you do that? When you do an ATL+/ it brings up a box that you type in a question. No close button (Escape only), you can't resize it, you can't move it. Copilot Chat is in it's own window, but not the Ask Copilot window. It's like the built the prototype for it and someone said "Ship it!". The rest of Visual Studio is very polished and great. Copilot, not so much.