“Our AI is called a ‘co-pilot’ because decisions are made by a human being”: Thomas Dohmke, CEO of GitHub

THE programmers Various tools are used today. From software that helps you visualize your work to machetes to remember commands, sit down and write the code It is a more accessible task thanks to the advancement of technology. But among all this, perhaps the most used platform is a site where there are millions of repositories: Git Hub.

The site is a website that began in 2007 as a collaborative work center to host open source projects which, today, is more of a portfolio where programmers store their work, both publicly and privately. It has grown so much that in 2018, Microsoft bought it for a whopping $7.5 billion.

In a context where the artificial intelligence (AI) Through any “tech” discussion, GitHub pushes a system called Co-pilot which assists programmers in the process of writing and debugging code. And it’s largely the challenge of its current CEO, thomas dohmke: explain how it works, what it can do, and most importantly, what it can’t do.

Dohmke passed through Buenos Aires this week and had an almost celebrity presence, visiting companies like Microsoft AND globaluniversity and even gave a student a stuffed animal Ottocat, the GitHub mascot (a half-octopus, half-cat creature). Her resume, to a large extent, explains the phenomenon: she is a senior manager, but keep planning.

From East Germany to GitHub

Born in the late 1970s, Dohmke grew up in the German Democratic Republic when the country was divided into a capitalist and a socialist faction. So, I used a clone of the KC 85a set of microcomputers from East Germany, found in his school’s geography lab.

He created his first programs in basic languagepure trial and error, with a few manuals that have crossed the wall and a perfect attendance of the “computer club” every Wednesday, a space where information was shared: the blood open source It was already running through his veins. He also dealt with the DOS operating system, with some friends, a kind of Monkey Islandthe classic graphic adventure game.

From then on, having studied computer engineering already in unified Germany, he tackled a number of his own projects and everything was “push”as they say in jargon: write source code.

It was a long time before he became the head of one of the most important programming sites, which today is essential for anyone active in the industry and functions as the LinkedIn of programmers.

Here is a chat with clarion on discussing artificial intelligence, copyright, learning to code, and more.

─What was the first thing you planned professionally?

─Software insurance. During the 1990s, when I was in my final years of high school, I developed a program with my brother (who was working as an insurance agent at the time). The program was like stacks of paper with all the insurance information: postcodes for houses, number plates for cars, and so on. we sent them floppy disks to around 500 insurance agents in Germany every three months.

─And what did they do with it?

─We scanned all that information with a OCR (text recognition system) to have the numbers in a database and build the program. We ripped records every day, we had like 1,500 3 and 1/2 disks. I’ve put together a version for the Friend and then for MS-DOS. It wasn’t until some time after that that we had Windows compatibility. It lasted until the late 2000s, when all the insurance companies had their own software and gave it away for free, because their business was selling insurance, not software. But this allowed me get my first job “without working”: I didn’t feel like working, I thought it was fun.

─ How would you explain programming to someone who doesn’t program?

─The thing to understand is that most web applications or programs today are written in what we call “programming language”, which is similar to a human language, but has a very specific vocabulary with a small number of words that can be written. A developer is someone who, after all, writes text files with code in different languages: Python, Ruby, Swift, Kotlin, etc., serving different needs. These files are called “repository”.

─And what role does GitHub play for software developers?

─With the invention of the Internet, open source has become increasingly important. Open source means I share my code source with anyone else. The contrast is with closed source code: I don’t give you my source code, it’s copyrighted. GitHub started out as a place to share open source work, but today it has both kinds of content: repositories open and closed source.

─From those beginnings until today the platform has changed a lot.

─Of course, GitHub also has more specific services and sells them to both small and large companies. Here in Argentina, for example, Free market use GitHub to build your own internal software. But the spirit remains: most companies use the 90% open source in its applications.

The artificial intelligence boom: welcome to the co-pilot

─AI has been around for decades but exploded in the media at the end of last year. Why?

─I think with ChatGPT it has become very obvious that AI is more than something to play with. The platform already has more than 100 million users, it’s the fastest “zero to a hundred” we’ve seen in history. And the best part is that you and any of us can use it – people have started to take notice the power of the tool. And this was leveraged by the alliance of Microsoft, which owns Github, with OpenIA, which owns ChatGPT, in the development of Copilot.

─What is Copilot and how does it work?

─Copilot try to predict what do you want to write next. Many times developers not only write code that executes business logic, but also code that tests that logic against really complex algorithms. With Copilot what you can do is, for example, write 20 or 30 lines of code that are simply synthesized with autocompletesimply by pressing the button TAB to “fill” what we call “ghost text”.

─With all the information you have on using Copilot, what have you found regarding productivity?

─Definitely improvement of productivity, as in many cases it frees the developer from doing repetitive work he already knows and leaves him more time for creative work. What we are seeing is that, on average, 46% of the code is written by Copilot. For some languages ​​like Java, the figure also reaches 60%

─Can you give a specific example of a Co-Pilot?

─Yes, of course. Before coming to Buenos Aires, I went through Brazil and gave a lecture entitled How to build an app in 18 minutes. I used Copilot to do the classic snake game! The one that was in the 1997 Nokia 6110. The first three minutes were chats: the remaining 15, the time it took me to program it. I think this demonstrates the power of AI and How can we take advantage of it?

─All information is taken from what people upload to GitHub. How does copyright work? Doesn’t Copilot “take” other authors’ code?

─The system works in such a way that if the code is longer than 150 characters, that code is compared to any code on GitHub and then basically search the database and try to find if there’s a code twin, regardless of license, doesn’t matter if it’s open or private. When Copilot finds a match, it won’t show the code to protect the author of the code, whether they are companies or individuals, to avoid copyright claims.

─Would you recommend using Copilot to someone who is learning to code, or is it better to use it later?

─It should be understood that these large language models based on ChatGPT, which is the basis of Copilot, They are not sensitive or intelligent. They can only create when you ask them to create something, and they do it within the limits you set for them. Co-pilot it’s like having a tutor which will help you when you get stuck, when you find yourself at a dead end. Or he will suggest how to solve a problem. However, we call it a co-pilot for a reason: the decisions are always ultimately made by the pilot. And that pilot is a human being.