AI pair programming

[Guest]

Has anyone used AI pair programming? 

Linus talked about it on wan show recently and I remember a lot of buzz going around with GitHub copilot. 

 

Has anyone used it? 

How is it in a large project?

What did you use? 

 

Did you like using it? 

Would you recommend it to an intermediate programmer? 

[Takumidesh]

yes I have; I used copilot in the beta and have been messing with it with chatGPT.

it's.. alright...
there are some moments of pure brilliance, and then moments of utter BS that set me down a weird rabbit hole.

chatGPT can be helpful as a rubber ducky, which is nice.

I am interested in trying codex out a bit, same GPT3, but focused on programming.
 

[Guest]

1 hour ago, Takumidesh said:

yes I have; I used copilot in the beta and have been messing with it with chatGPT.

it's.. alright...
there are some moments of pure brilliance, and then moments of utter BS that set me down a weird rabbit hole.

chatGPT can be helpful as a rubber ducky, which is nice.

I am interested in trying codex out a bit, same GPT3, but focused on programming.
 

Would you say it's worth the hassle or no real benefit? 

 

Or gets in the way? 

[Takumidesh]

I often found copilot to get in the way. Once you start getting more complex then getters and setters or other boilerplate, it starts to go off the rails more than its helpful.

it may be better depending on the type of code being written though.

[Guest]

19 minutes ago, Takumidesh said:

I often found copilot to get in the way. Once you start getting more complex then getters and setters or other boilerplate, it starts to go off the rails more than its helpful.

it may be better depending on the type of code being written though.

May I ask what type of projects and code you were writing with it? 

[Takumidesh]

8 minutes ago, fpo said:

May I ask what type of projects and code you were writing with it? 

Yea heavy math, .NET WPF desktop apps for processing data from industrial equipment (specifically metrology equipment).

lot's of interfacing with PLCs and embedded controllers, Ingesting 2-3 million data points (2d & 3d point clouds) and processing them, doing stuff like lines of best fit and really heavy list manipulation.

[Guest]

2 minutes ago, Takumidesh said:

Yea heavy math, .NET WPF desktop apps for processing data from industrial equipment (specifically metrology equipment).

lot's of interfacing with PLCs and embedded controllers, Ingesting 2-3 million data points (2d & 3d point clouds) and processing them, doing stuff like lines of best fit and really heavy list manipulation.

Ah yeah, definitely not something I'd imagine GitHub copilot to be trained on. 

[Takumidesh]

there are moments that it has surprised me, but unfortunately, it is difficult to trust the math that it generates, so the assumed time savings are lost when you need to figure the math out anyway to be sure.

[gstan]

I've been using Copilot for a couple months and for web related projects or working with open-source software, it's ok-ish. It usually understands the context and sometimes provides a working solution but most of the time, you will definitely alter the code it provides.

[Guest]

59 minutes ago, gstan said:

I've been using Copilot for a couple months and for web related projects or working with open-source software, it's ok-ish. It usually understands the context and sometimes provides a working solution but most of the time, you will definitely alter the code it provides.

Is this for bug fixes or writing new features? 

[markdamon90]

Dude, this ChatGPT is seriously smart and always comes in handy.

[Franck]

On 12/6/2022 at 10:02 PM, fpo said:

Has anyone used AI pair programming? 

Linus talked about it on wan show recently and I remember a lot of buzz going around with GitHub copilot. 

 

Has anyone used it? 

How is it in a large project?

What did you use? 

 

Did you like using it? 

Would you recommend it to an intermediate programmer? 

I never used CoPilot but to answer as the first question if i have used AI paired programming yes i did.

 

Has anyone used it ?

I've been using the new IntelliSense AI inside Visual Studio for a bit more than a years now.

It works great. It does take 3 months to fully kick in as it first learn your

habits and usual context but when it starts it is strong.

 

How is it in a large project?

I use it in small and large projects but for reference as people do not always have the same scale for me small mean little library of maybe hundred or less lines of code and large means a 4.5-5 gb of code. The more code you have in the project the more accurate it is as it use all your code to guess what you need.

 

What did you use? 

What i use 99% of the time is the autocomplete feature and line suggestions

 

Did you like using it? 

Yes, it does save little time here and there and the suggestion sometime hinted me alternative and more efficient way of doing little things here and there. It's subtle but it does work very very well.

 

Would you recommend it to an intermediate programmer? 

Yes as long as the programmer does not take it for granted as the true solution. It is a tool that help with little push most of the time like correcting the parentheses location, setting eol character. It's solutions also sometime ping the obvious as "hey maybe you have a mistake here" when you have a long set of else if that one of them is written differently it will warn you that it's different than the other and that you might have wanted to do something else similar. I will make them think twice about what they do and making someone question himself make that person become better at his job.

[Guest]

42 minutes ago, Franck said:

I never used CoPilot but to answer as the first question if i have used AI paired programming yes i did.

 

Has anyone used it ?

I've been using the new IntelliSense AI inside Visual Studio for a bit more than a years now.

It works great. It does take 3 months to fully kick in as it first learn your

habits and usual context but when it starts it is strong.

 

How is it in a large project?

I use it in small and large projects but for reference as people do not always have the same scale for me small mean little library of maybe hundred or less lines of code and large means a 4.5-5 gb of code. The more code you have in the project the more accurate it is as it use all your code to guess what you need.

Yeah when I look of large project I think of incredibly huge sell for money project worked on by at least 10 full time people. Or something within the realm. 

42 minutes ago, Franck said:

What did you use? 

What i use 99% of the time is the autocomplete feature and line suggestions

 

Did you like using it? 

Yes, it does save little time here and there and the suggestion sometime hinted me alternative and more efficient way of doing little things here and there. It's subtle but it does work very very well.

 

Would you recommend it to an intermediate programmer? 

Yes as long as the programmer does not take it for granted as the true solution. It is a tool that help with little push most of the time like correcting the parentheses location, setting eol character. It's solutions also sometime ping the obvious as "hey maybe you have a mistake here" when you have a long set of else if that one of them is written differently it will warn you that it's different than the other and that you might have wanted to do something else similar. I will make them think twice about what they do and making someone question himself make that person become better at his job.

May I ask what technologies you used with it? 

Java, C++? 

HTML? 

 

Which visual studio is this in? I should like to give it a try. 

Is it called "intellisense ai?" 

[Franck]

27 minutes ago, fpo said:

Yeah when I look of large project I think of incredibly huge sell for money project worked on by at least 10 full time people. Or something within the realm. 

May I ask what technologies you used with it? 

Java, C++? 

HTML? 

 

Which visual studio is this in? I should like to give it a try. 

Is it called "intellisense ai?" 

Mainly C# for work but C++ at home.

This is part of Visual Studio 2022. It is part of Intellisense which has been enhanced with AI that was trained with information gathered from developper around the world running Visual Studio 2019 and it also used public github for the learning model. It is even more powerful if you use comments as it can understand what you are trying to achieve and it will suggest full code for it as you go based on a single comment.

 

[igormp]

On 12/7/2022 at 12:02 AM, fpo said:

Has anyone used it? 

Yes, I've been using copilot for over an year now.

On 12/7/2022 at 12:02 AM, fpo said:

How is it in a large project?

Works pretty nice for simpler stuff, and it's able to suggest some basic patterns that you are currently writing.

On 12/7/2022 at 12:02 AM, fpo said:

What did you use? 

Copilot with VSCode on Python, C, Cpp codebases, with tons of bash, yaml and dockerfiles in between.

On 12/7/2022 at 12:02 AM, fpo said:

Did you like using it? 

Would you recommend it to an intermediate programmer? 

Yes, can recommend it, I feel like I can get stuff done faster with it.

[Takumidesh]

On 1/14/2023 at 2:49 PM, markdamon90 said:

Dude, this ChatGPT is seriously smart and always comes in handy.

The problem is when you are doing serious work of moderate to severe complexity.

ChatGPT is like a compulsive liar, it might be right most of the time, but because it isn't 100% it ultimately makes it hard to trust ANY of it.

If you are asking questions that you don't know the answer already, it can effectively spout BS at you in a believable enough way that you can't tell that it is wrong.

If you ask it a question you do know, there is not really much value add, since it confirming your knowledge doesn't actually prove that what you thought was correct is in fact correct. It is not an authority on a topic and there exists no way to challenge it, other than having a known definite correct answer to compare.

 

The difference between a tool like ChatGPT and stack overflow, is that there is peer vetting. if someone provides misinformation it can be corrected, or at least challenged.

Anything chatGPT tells you, you have to take it at face value, or just google after it and since you can't trust the face value message, you end up researching it 'the old way' anyway.

 

I would probably get chewed out if I told my boss I used chatGPT to solve a real problem.

 

[Guest]

Yeah @Takumidesh@markdamon90

I tried using chat gpt and it came across extremely limited. 

I have zero fear for terminators rising up soon. At least in ... Intelligent form. Otherwise it's like facing multiplayer bots.