Up next in 10
LIGHTNING FAST Code From Artificial Intelligence
7K views
Feb 6, 2024
"LIGHTNING FAST Code From Artificial Intelligence" is an intriguing video showcasing the capabilities of artificial intelligence (AI) in generating code swiftly. The video likely demonstrates how AI-powered tools can automate and expedite the coding process, reducing the time and effort required to develop software. It may feature examples of AI algorithms generating code snippets based on natural language descriptions or automatically completing repetitive tasks in programming workflows. Additionally, the video might discuss the potential impact of AI on software development, including improved productivity, code quality, and the democratization of coding skills. By highlighting the efficiency and speed of AI-generated code, the video aims to showcase the transformative potential of AI in the field of software engineering.
View Video Transcript
0:00
GitHub Copilot, your AI pair programmer or is it
0:03
Copilot is more than just an autocomplete plugin. GitHub Copilot is an open AI system called Codex
0:09
It's a way to use AI to generate code snippets, tests, and other boilerplate
0:14
And if you don't like its suggestions, then don't worry because it generates an entire list of different suggestions. Basically, you just install an extension
0:21
in your favorite editor and you're good to go. It works across a ton of different
0:24
languages and frameworks and it works especially well if it's one of the more popular languages
0:29
like JavaScript, Java, Python. Easy peasy. And who needs programmers anyways? I'm just curious, have you used Copilot before? I'd love to hear your experience
0:37
so drop it in the comments. I do read them. So how does GitHub Copilot work? Well
0:41
you drop a comment describing what you want to have happen and it offers you suggestions based
0:45
on billions of lines of public code that was used to train its AI. It's so easy that in a
0:51
matter of minutes, it created an application that I was able to deploy to the app store and it's
0:55
already making me a ton of money. Looks like I'll be out of a job soon, but at least I'm on my way
0:59
to becoming a millionaire. You probably shouldn't even think of becoming a programmer because you're
1:03
just too late. Okay, so I am totally kidding there. It did not build me an application like
1:08
that and I'm not really worried about artificial intelligence destroying the future of software
1:13
development. But I know a lot of people out there are because I get asked it a lot. What does concern
1:17
me is a massive wave of lawsuits. But before we get into that, we have to talk about two important
1:22
concepts that you must understand before everything else makes sense. First, there are a
1:27
ton of different ways companies protect trade secrets and code. It can be as simple as keeping
1:31
all source code private, but valuable information is usually protected through trade secrets
1:37
copyrights, and patents. I'm not an attorney and so this is not meant to be legal advice
1:41
or a comprehensive description of how this stuff works. I am going to simplify it for discussion
1:47
sake. Basically, copyrights protect the actual text that you write, the images, the content that
1:52
you create, but it does not protect the actual ideas or concepts. And it's fairly easy to get
1:57
copyright protection in the United States. Patents are much more involved and they can take years to
2:02
secure and usually cost thousands of dollars. But patents protect the new and unique ideas
2:07
the systems, processes, and stuff like that. Whereas the copyright is just going to protect
2:12
the actual source code you write as it is on the page. The second thing that you need to understand
2:16
is licensing. If you want to use software that has been created by someone else, then you're
2:20
going to have to ask permission. And this permission is called getting a license. For
2:24
example, if you pay for software, you don't actually own the software, you're just paying
2:28
for a license to use it. And that license is going to say things like what you can and can't do with
2:33
it. For example, how many users can access it and how many computers it can be on. Now some
2:38
companies tightly control their code and others decide that they're just going to share it as
2:42
open source. But not all open source projects are the same. Some licenses will let you do pretty
2:47
much whatever you want. You can copy it, build off of it, whatever you can make money with it
2:53
no strings attached. But other license types will let you use their code inside your code for free
2:57
as long as the software product is free. If you end up making software for profit
3:02
then you're going to end up having to pay them for a commercial license. If this is helpful
3:06
smack that like button. But there are some licenses that will really bite you in the butt if you
3:10
aren't careful. For example, some open source licenses may say that if you use their code
3:14
in your project, that you're going to then have to make your software open source. And companies
3:19
don't want to do this. They don't want to include this kind of code, because they can spend millions
3:23
of dollars developing software. And they don't just want some competitor to walk in and be able
3:28
to use their software for free. Because of some open source requirement. This last type of license
3:33
is really important to what we're about to discuss. You see, GitHub Copilot was trained on
3:38
billions of lines of public code stored in public repositories. If you aren't familiar with the term
3:44
repositories, that's just another name for where all of the code for a software project is stored
3:49
Anyways, each of these projects likely has a license file where the owner can specify
3:53
the terms for using the code. In GitHub, you'll find a ton of different licenses
3:58
including the ones that can bite you. This is what really worries me about GitHub Copilot
4:03
And I did a quick search to get a vibe of how other devs feel about this. And the concern is
4:09
not just with me, you see, because Copilot is trained on data from a ton of different sources
4:14
you could pretty easily argue that anything Copilot creates is a derivative of those
4:20
projects. And you must follow the terms and licensing for those projects. But how do you
4:25
know where the code came from? You just don't, it offers you code suggestions, and you don't
4:31
really have a reliable way to trace the origins. I mean, you could take that snippet and you could
4:36
drop it into Google and see if anything similar shows up. But let's see what Copilot has to say
4:40
about this. On their website, there's a section titled does GitHub Copilot recite code from the
4:46
training set, you could pause and read the whole paragraph because it's very interesting. But the
4:51
short answer is no, it claims the code is uniquely generated, except for about point 1% of the time
4:59
when the snippet is verbatim. Now, that might seem like a low risk. But let's be real here
5:04
depending on where that point 1% comes from, you could be talking a multi million dollar lawsuit
5:10
some lawyer shows up at your desk asking you why you stole code from a competitor. What's your
5:15
answer? It wasn't me, it was Copilot. Well, we can't prove that but your personal name is on
5:21
the git blame for that line of code. So you're in deep doo doo unless you subscribe, then I'll
5:26
forgive you no matter how bad you write code. Or what about this other section that talks about
5:31
personal data? Now personal data should be pretty easy to spot in the suggestion. But as a GitHub
5:36
repository owner, I would be even more concerned about what is being stored in my repo. Not that
5:42
you should be storing any kind of personal information there anyways, that's just bad
5:46
practice. Apparently, some companies are concerned enough by Copilot that they're pulling their
5:51
projects from GitHub. But is that the right answer? For starters, I would personally encourage you to
5:56
keep any important and sensitive repos private anyways, instead of making them public. But I
6:02
think this problem is bigger than GitHub and even Copilot. Now I'm a realist and like it or not
6:08
artificial intelligence is going to play a big part of the world we live in. And I think that AI
6:12
is going to be a major disruptor when it comes to challenging the way our society handles patents
6:17
and licensing. I mean, it could completely change the way that we protect trade secrets, how we
6:22
write and interact with our code, as well as handle attributions. Sure, it's probably fun to
6:26
experiment with Copilot for personal projects that you don't plan to commercialize. But until these
6:31
patent laws change and catch up with this use of artificial intelligence, there's just no way in
6:36
heck that I would use it for writing code for my employer. It's just not worth the liability. If I
6:43
write code, I know what I've written, and which libraries I have leveraged, and I will be able to
6:49
explain that. I will know the exact origin of that code. And if you think that this is bad
6:54
then you should watch this video on how one company is using AI to destroy the job interview
6:59
process as we know it. Lates
#Development Tools
#Intellectual Property
#Legal
#Open Source
#Other