AI-Powered GitHub Copilot Leaves Preview, Now Costs $100 a Year (techcrunch.com) 36
It was June 29th of 2021 that Microsoft-owned GitHub first announced its AI-powered autocompletion tool for programmers — trained on GitHub repositories and other publicly-available source code.
But after a year in "technical preview," GitHub Copilot has reached a new milestone, reports Info-Q: you'll now have to pay to use it after a 60-day trial: The transition to general availability mostly means that Copilot ceases to be available for free. Interested developers will have to pay 10 USD/month or $100 USD/year to use the service, with a 60-day free trial.... According to GitHub, while not frequent, there is definitely a possibility that Copilot outputs code snippets that match those in the training set.
Info-Q also cites GitHub stats showing over 1.2 million developers used Copilot in the last 12 months "with a shocking 40% figure of code written by Copilot in files where it is enabled." That's up from 35% earlier in the year, reports TechCrunch — which has more info on the rollout: It'll be free for students as well as "verified" open source contributors — starting with roughly 60,000 developers selected from the community and students in the GitHub Education program... One new feature coinciding with the general release of Copilot is Copilot Explain, which translates code into natural language descriptions. Described as a research project, the goal is to help novice developers or those working with an unfamiliar codebase.
Ryan J. Salva, VP of product at GitHub, told TechCrunch via email... "As an example of the impact we've observed, it's worth sharing early results from a study we are conducting. In the experiment, we are asking developers to write an HTTP server — half using Copilot and half without. Preliminary data suggests that developers are not only more likely to complete their task when using Copilot, but they also do it in roughly half the time."
Owing to the complicated nature of AI models, Copilot remains an imperfect system. GitHub said that it's implemented filters to block emails when shown in standard formats, and offensive words, and that it's in the process of building a filter to help detect and suppress code that's repeated from public repositories. But the company acknowledges that Copilot can produce insecure coding patterns, bugs and references to outdated APIs, or idioms reflecting the less-than-perfect code in its training data.
The Verge ponders where this is going — and how we got here: "Just like the rise of compilers and open source, we believe AI-assisted coding will fundamentally change the nature of software development, giving developers a new tool to write code easier and faster so they can be happier in their lives," says GitHub CEO Thomas Dohmke.
Microsoft's $1 billion investment into OpenAI, the research firm now led by former Y Combinator president Sam Altman, led to the creation of GitHub Copilot. It's built on OpenAI Codex, a descendant of OpenAI's flagship GPT-3 language-generating algorithm.
GitHub Copilot has been controversial, though. Just days after its preview launch, there were questions over the legality of Copilot being trained on publicly available code posted to GitHub. Copyright issues aside, one study also found that around 40 percent of Copilot's output contained security vulnerabilities.
But after a year in "technical preview," GitHub Copilot has reached a new milestone, reports Info-Q: you'll now have to pay to use it after a 60-day trial: The transition to general availability mostly means that Copilot ceases to be available for free. Interested developers will have to pay 10 USD/month or $100 USD/year to use the service, with a 60-day free trial.... According to GitHub, while not frequent, there is definitely a possibility that Copilot outputs code snippets that match those in the training set.
Info-Q also cites GitHub stats showing over 1.2 million developers used Copilot in the last 12 months "with a shocking 40% figure of code written by Copilot in files where it is enabled." That's up from 35% earlier in the year, reports TechCrunch — which has more info on the rollout: It'll be free for students as well as "verified" open source contributors — starting with roughly 60,000 developers selected from the community and students in the GitHub Education program... One new feature coinciding with the general release of Copilot is Copilot Explain, which translates code into natural language descriptions. Described as a research project, the goal is to help novice developers or those working with an unfamiliar codebase.
Ryan J. Salva, VP of product at GitHub, told TechCrunch via email... "As an example of the impact we've observed, it's worth sharing early results from a study we are conducting. In the experiment, we are asking developers to write an HTTP server — half using Copilot and half without. Preliminary data suggests that developers are not only more likely to complete their task when using Copilot, but they also do it in roughly half the time."
Owing to the complicated nature of AI models, Copilot remains an imperfect system. GitHub said that it's implemented filters to block emails when shown in standard formats, and offensive words, and that it's in the process of building a filter to help detect and suppress code that's repeated from public repositories. But the company acknowledges that Copilot can produce insecure coding patterns, bugs and references to outdated APIs, or idioms reflecting the less-than-perfect code in its training data.
The Verge ponders where this is going — and how we got here: "Just like the rise of compilers and open source, we believe AI-assisted coding will fundamentally change the nature of software development, giving developers a new tool to write code easier and faster so they can be happier in their lives," says GitHub CEO Thomas Dohmke.
Microsoft's $1 billion investment into OpenAI, the research firm now led by former Y Combinator president Sam Altman, led to the creation of GitHub Copilot. It's built on OpenAI Codex, a descendant of OpenAI's flagship GPT-3 language-generating algorithm.
GitHub Copilot has been controversial, though. Just days after its preview launch, there were questions over the legality of Copilot being trained on publicly available code posted to GitHub. Copyright issues aside, one study also found that around 40 percent of Copilot's output contained security vulnerabilities.
Private API keys (Score:4, Funny)
Re: (Score:1, Troll)
It rips off open source code and strips the licence. They basically want $100 to help people violate open source licences.
Re:Private API keys (Score:5, Interesting)
It also seems highly speculative to consider an AI model based on lots of other people's work (with no regards to licensing incompatibilities) to itself not be a derived work. It hasn't been tried in court yet, but I wouldn't be betting my business on it.
Re: (Score:2)
It also seems highly speculative to consider an AI model based on lots of other people's work (with no regards to licensing incompatibilities) to itself not be a derived work. It hasn't been tried in court yet, but I wouldn't be betting my business on it.,
Copyright only applies to creative aspects of a work not 'merely functional' aspects. Generic function names are by definition not creative, as are standard algorithms and code expressions. Since these are the only things that the models can do, there isn't likely to be a risk of copyright violation.
Re: (Score:2)
Interesting this was modded troll. Either there are some shills in the house or my personal troll is back.
Re: (Score:2)
That's pretty much it. What else needs to be said?
$100 a year (Score:2)
How much of this fee goes towards insurance against the licensing lawsuits that are bound to occur?
Re: (Score:3)
Higher-level language? (Score:1)
It seems that AI-assistance is just the next level of programming language.
They should just go the whole hog and let the thing program in assembler while taking commands in human language. :P
Re: (Score:2)
Good luck spotting the subtle bug when it inserts half a page of AI-generated code for you.
Re: (Score:2)
You write a test for it like for any other code.
And if it "bugs" you step though it with a debugger.
Good luck finding the bugs I introduce by typing manually ...
Oops: there is no difference.
Re: (Score:2)
The more you rely on AI generated code, the closer that foot is to the ground.
Re: (Score:2)
If the AI would not help wring the code that fits to my mental model: it would be completely useless, or not?
Perhaps I have to install it and test it ...
Re: Higher-level language? (Score:2)
Compilers are so good at optimizing these days that it makes very little sense to write an entire program in assembly. Maybe do a bit of inline assembly here and there to replace a hot spot or two, but the odds of needing to do that at all are slim to none. The best case I can think of is when you want tail recursion but the compiler just isn't having it.
Re: (Score:2)
Compilers aren't magic, you know. Even the best compiler won't significantly improve the lazy / incompetent garbage I've seen, but all the kids seem to think that it doesn't matter what they write, the enchanted compiler will somehow make everything better.
This blind faith people have in modern tools is irresponsible. You should be extremely distrustful of anything "helpfully" produced for you. After all, as our tools have "improved", software quality has only gotten worse.
Another_really_ bad idea (Score:4, Insightful)
The fuckups that cannot code (https://blog.codinghorror.com/the-nonprogramming-programmer/) will probably blame all their mistakles which come from sheer incompetence on that tool.
Re: (Score:2)
And Microsoft is doing the best possible filtering of their audience. Since only a dumbass would pay for this feature, they are guaranteed a user base of people who won't know what's wrong with it.
Re: (Score:2)
Well, nobody ever blamed Microsoft to be bad at marketing. They are just bad at everything else they do.
Amazing... (Score:3)
Pay 100$ to train our models for us. They gave it away for free long enough to have people teach it the basics, and now people must pay for the privilege of teaching their replacement more.
Re: (Score:2)
yeah don't really see a problem with that? i mean, i'm not going to use it but you have to give it to them, well played.
Comment removed (Score:4, Insightful)
Re: (Score:2)
And the copyright goes to ... (Score:2)
So, if you use this "Copilot" to help write the code, does Microsoft/GitHub get a copyright credit as well, or just you?
Re:And the copyright goes to ... (Score:5, Insightful)
no, it's a tool, your work is still yours (your boss'). of course some bat crazy moron can come up with such an idea and, in the bottom of our hearts we could not blame him given the aberrant clusterfuck and intellectually dishonest exercise of power intellectual property law actually is, so yes, waiting for some moron suing another moron just because of smart autocomplete in 3, 2, 1 ...
Re: (Score:3)
If the smart autocomplete turns out to lift lines verbatim from code with an incompatible license then guess what? It doesn't matter whose fault it is, you've still got that contamination in your code, opening up the possibility for a lawsuit that is expensive even if Microsoft turns out to be liable.
Re: (Score:3)
Copying single lines of code, regardless what license the other one is: does not fall under any copyright protection.
Otherwise no one would be able to read something up and reuse it for his problem. There are plenty of things that do not fall under copyright protection anyway. Or do you think an AVL tree is copyrighted by Knuth or Wirth?
Re: (Score:2)
Copying single lines of code, regardless what license the other one is: does not fall under any copyright protection.
"copyright infringement can be made out where only a small proportion of source code [lexology.com] is taken from a piece of software, where that source code constitutes a functionally significant part of the software."
If that piece of software was very small to begin with, a single line can be functionally significant. And you have no idea where it's getting its code. Is it copying it? Is it writing it? Nobody knows.
Re: (Score:3)
Perhaps you should read the court ruling?
a single line can be functionally significant.
Every line is functionally significant. Or it would not be there - unless it is dead code.
Re: (Score:1)
Copying single lines of code, regardless what license the other one is: does not fall under any copyright protection.
That's simply not true. The requirement is creative expression, which is a pretty low-bar that many individual lines of code can easily clear.
Or do you think an AVL tree is copyrighted by Knuth or Wirth?
You can't copyright an algorithm. However, you can copyright the expression of an algorithm (an implementation) or a description of an algorithm. I can assure you that Knuth's description in The Art of Computer Programming: Volume 3 has copyright protection. The same is true of Wirth's description in Algorithms and Data Structures.
You should really take the time t
Are you tired, worn out? (Score:5, Funny)
*Please note GitHub Copilot will not attend meeting for you.
Just what we need (Score:2)
More mystery meat code that the programmer most likely wouldn't really be able to understand and debug.
Don't forget the fun with shitting in the pool with malicious code.
This seems perfect for those "Google Programmers" (Score:2)