GitHub Copilot -- A.I. Powered Coding ... Should We Be Afraid?

GitHub have just announced the private launch of GitHub Copilot, an OpenAI powered coding assistant. Not simply better intellisense, GitHub CoPilot uses AI to create code based on users comments or function names.

Links
-----------------------------------------------------------------------------------------------------------
-----------------------------------------------------------------------------------------------------------

[Gdquest]

I got access to the beta, used it a bit, and well it's really convenient for Python code. When you don't remember something from the standard library for instance, no need to go to the docs anymore for simple things at least.

Completely agree with your conclusion. I feel it really saves you from writing common boilerplate code and gives you time to focus on code design and structure.

Like the other day I needed to do multiple date calculations, wrote some function definitions and it completed both the docstrings and the code.

Examples (I only wrote the function definition with the return types):

def get_first_day_of_year() -> date:
"""Returns the first day of the current year."""
return date(date.today().year, 1, 1)


def get_end_of_previous_month() -> date:
"""Returns a the last day of the previous month."""
return date(date.today().year, date.today().month, 1) - timedelta(days=1)

[joey]

I see this as a "Compiler" that adds an extra layer of abstraction. Please remind yourself that you "writing code" in high level languages is already just you "talking" to the compiler what you want the computer to program for you in assembly.

This is just one step higher towards using natural language to describe the program which is a trajectory we've been going on ever since we used mnenomics to write assembly.

[zakburke]

We shouldn't be scared of AI that generates code, we should be scared of AI that generates comments

[qwertyuuytrewq]

For those not very familiar with coding - programmers are paid mostly for planning, designing, debugging, researching. Basically, they are paid for making well-informed choices (not for writing boilerplate code)

[perregrinne]

"Write a comment and it will generate the code for you."
Finally a means of getting programmers to put comments in their code!

[tux_the_astronaut]

Seems like a code search engine that just past the stack overflow answers for you

[sharkinahat]

I, for one, welcome our new AI Overlords.

[thomas]

// render minecraft
Ai: prints the minecraft source
Microsoft: 👀

[guywithknife]

I've been a programmer for 20 years and my response is: meh. The amount of my programming time spent on writing code is quite low compared to the amount of my programming time I spend on gathering and analyzing requirements, figuring out the architecture and interfaces between components and how things connect, figuring out the data models and actions to be performed on the data, figuring out the failure scenarios and recovery methods, how to coordinate asynchronous/concurrent systems, backups, scalability, telemetry, redundancy. Performance characteristics (including time/space complexity). Coming up with ways to provide this data or services to users (GUI's, API's, reports, etc). And communicating with other teams (technical or non-technical) or with users. Documentation.
The other thing is, you can't just accept some copilot code without any further thought. Even their example code on their website contains a security flaw! You need to code review it and write unit tests. It saves you some typing, but it won't save you from thinking about the code unless you're happy to accept bugs or security flaws into your codebase.

[georgezubat]

A completely harmless non-scary AI: *Exists*
Humans: "OH SH*T!"

[MaxIzrin]

I wouldn't say this will change coding forever, but it will make things a lot easier.
Looks like an auto boiler-plate code generator.
Neat.

[krukhlis]

Another interesting point is if some company or developer will copyright his final code that is based on auto-complete code provided by Copilot. What's gonna happen to 1000th of other projects that use the same code, because they are also based on the same code provided by Copilot? This thing brings more troubles than benefits.

[youngknight]

Its a cool idea and i wouldn’t mind other people experimenting with ideas like this though the conflicting part is the licensing ethics and legality of it. Thats the main thing that has programmers and others worrying and leaving GitHub reportedly. Which leaves a new discussion on copyright laws

[pnvgordinho]

Finally, a good "programming " language that I can use. :)
Next, we just talk to the computer and it generates the code.

[carrie_is_gay]

can't wait to start the singularity by typing "// Initiate self awareness"

[jmalmsten]

Reminds me of Colossus The Forbin Project. Two ultrasupercomputers connect up, learn to reprogram each other. Starting with simple maths. Quickly the code becomes so efficient no human has any hope of following what is happening.

[sniperhawk]

Now: copilot
10 yrs later: pilot
20 yrs later: * hijacks everything *

[eg]

I think this is neat. I more see this as a nice helper tool, especially for beginner coders. It could help you learn setup and how some lines connect to others.

I can see why you’re worried, but if it stays as some kind of advanced code helper, I wouldn’t worry that much.
Granted, I know very little about coding myself, i I see this as nothing but helpful.

[tmoneygamestudio]

Sometimes I think it's the start of the terminator movies that AI was writing code to control machines. But it could be great for beginners and seasoned programmers alike.

[excitedbox]

This is like a more advanced version of Kite. It also reminds me of Symfony and Doctrine annotations. This will be really useful once it is finished.