When software developer Nikolai Avteniev got a preview of Microsoft’s Copilot coding assistant in 2021, he quickly saw the potential.
The assistant was developed on Microsoft’s GitHub coding platform and is based on a version of OpenAI’s generative artificial intelligence, but it’s not perfect and can sometimes make mistakes. But Avteniev, who works at ticket seller StubHub, was surprised to see how neatly it could complete a few lines of code with just a few hints. All he has to do is press the Tab key, and Copilot fills in the rest.
“Instead of 15 keystrokes, we had three keystrokes,” he recalled recently. “That little increase in speed was nice.”
Three years later, GitHub’s Copilot incorporates the latest version of OpenAI’s GPT-4 technology and can do more, including answer engineers’ questions and convert code from one programming language to another. As a result, assistants are responsible for writing an increasing proportion of the software and even programming the company’s critical systems.
Along the way, Copilot is slowly revolutionizing the working lives of software engineers, the first professional group to collectively use generative AI. Microsoft says Copilot has attracted 1.3 million customers so far, including 50,000 businesses ranging from small startups to the likes of Goldman Sachs, Ford and Ernst & Young. Engineers say Copilot saves them hundreds of hours each month by handling tedious and repetitive tasks, freeing them up to focus on tougher challenges.
GitHub, which was acquired by Microsoft for $7.5 billion in 2018, dominates the market and is betting that Copilot has the artificial intelligence capabilities to take on rival services including Tabnine, Amazon’s CodeWhisperer and Google-backed Replit Ghostwriter. GitHub’s AI assistant is also a beta test for many other Copilots that Microsoft is integrating into Office, Windows, Bing and other lines of business.
Like artificial intelligence, GitHub Copilot has limitations. Developers say it sometimes pulls out-of-date code, provides useless answers to questions, and generates suggestions that are buggy or may violate copyrights. Because the tool is trained on public and open code repositories, engineers run the risk of duplicating security issues or injecting new ones into their work, especially if they blindly accept Copilot’s recommendations.
GitHub emphasizes that the tool is an assistant, not a replacement for human programmers, and puts the onus on customers to use it wisely. GitHub CEO Thomas Dohmke said strong guidelines are needed to prevent lazy programmers from simply accepting Copilot’s advice. He expressed confidence that engineers would stay honest with each other.
“The social dynamics of the team will ensure that those who cheat by accepting code too quickly and not actually going through the process defined by the team, that code will not go into production,” he said in an interview.
Generative AI is the latest in a series of innovations that have transformed computer coding over the years. Over the last century, program compilers have accelerated software development by quickly translating commands into 1s and 0s that computers can understand. Recently, Linux has popularized open source coding, allowing programmers to leverage each other’s work rather than writing everything from scratch.
Coding assistants like GitHub’s Copilot could be even more revolutionary, as generative AI has the potential to automate a lot of the work software engineers currently do.
For now, this mostly makes them more efficient. StubHub’s Avteniev, who also teaches software engineering at City College of New York, said Copilot’s predictive capabilities can help programmers stay “fluid” because they no longer need to stop and look for information. Avteniev has been coding for over 20 years, but even he sometimes forgets programming languages, forcing him to waste time Googling them. “Copilot saves you from having to exit your current coding process,” he said. “Even if it creates gibberish, it’s still easier to accept what it does and then correct it yourself.”
Aaron Hedges, a developer for over 15 years, was burned out before Copilot arrived. Hedges works for ReadMe, a startup that helps companies create technical descriptions of their application programming interfaces (APIs). Like Avteniev, he makes good use of Copilot’s autocomplete feature. “Because I’m a fairly senior engineer, I can look at it and say, ‘Oh, yeah, that looks good.'” He also likes that he can ask questions without leaving his programming window. “I don’t have to go away and open a browser, which could cause real damage,” he said.
The $10 monthly Copilot subscription is a bargain, and Hedges is willing to pay for it himself. In his spare time, he builds websites for Dungeons & Dragons fans. With a toddler and another baby on the way, downtime is at a premium. “The two hours I get to code by myself at night are really important to me,” he said. “The more efficient I can be, the better.”
There are few tasks more tedious than debugging software, a process that can take up 50% of an engineer’s time. Figma helps developers design application or website interfaces, and the company says Copilot can create defect testing programs in minutes instead of hours. “This is the true value of artificial intelligence,” said Abhishek Mathur, the company’s vice president of engineering. “It won’t replace our jobs, but it will free up our time to develop creative solutions.”
Some companies are beginning to deploy Copilot to create code for critical systems. Brewer Carlsberg uses it to code existing tools that help salespeople plan, prepare, and record sales calls. Chief Information Officer Sarah Haywood said that given Copilot’s limitations, the brewer uses its own quality assurance process to check that the code it creates works as expected. Eventually, she said, companies will be able to outsource this task as well. “Over time, people will build more trust in AI,” she said. “I don’t think we should scrutinize everything AI does, otherwise we’re not really adding any value.”
To assess the accuracy of the technology, Canada’s University of Waterloo published an experiment last year. The researchers collected a dataset consisting of code snippets with known flaws and the fixes for those bugs. The researchers prompted Copilot to create these precise snippets to see if it would spit out defective versions. The assistant copied a defective version 33% of the time, less frequently than humans. In a quarter of the cases, the AI outputs fix code. First officers are generally better at avoiding basic mistakes than more complex ones, said Mei Nagappan, one of the study’s authors and a computer science professor at the school.
“The analogy here is that we are now in the era of driver assistance but not yet in the autonomous driving phase,” he said.
Software engineers can be slow to change work habits. Many people welcome Copilot but worry about relying too much on it. A recent GitHub-funded study found that developers accepted assistant advice only 27% of the time.
Engineers are also quick to blame the co-pilot if something goes wrong. When Etsy’s website briefly crashed last October and December, some of the company’s developers blamed Copilot for the outage. Etsy confirmed the incidents but denied Copilot was responsible. “While we certainly understand that engineers may discuss how Copilot theoretically played a role in an outage or issue, we have zero evidence that the tool actually caused any customer-facing impact,” a spokesperson said.
Copilot is expected to improve significantly in the coming years. GitHub has rolled out enhancements, including an enterprise version that can answer questions based on customers’ own programming code, which should help new engineers get up to speed and enable veteran programmers to work faster. In the coming months, GitHub will also allow engineers to use their employer’s own code base to help automate the programs they are developing. This will make the generated code more customized and useful.
GitHub can’t sit back and wait. At least a dozen startups are looking to disrupt the market. Some are taking advantage of new models that greatly increase the amount of information code assistants can quickly leverage, making it easier for them to generate entire programs. “An AI programmer who can see all of the code will be able to make better decisions and write more coherent code than an AI programmer who can only see the code through a paper towel roll (a small amount at a time),” said investor and said Nat Friedman, former CEO of GitHub.
Friedman is backing a startup called Magic AI, which plans to create “superhuman software engineers.” Meanwhile, Cognition AI, backed by Peter Thiel, is developing an assistant that can handle software projects independently. Princeton University this month released an open-source model of an artificial intelligence software engineering agent, and new startups seem to be popping up every week.
In interviews, few programmers expressed concern that artificial intelligence would replace them. They say that, as in many industries, automation will free up their time to focus on more challenging and interesting tasks. But Jen-Hsun Huang, chief executive of Nvidia Corp., the hottest artificial intelligence chip maker, is less sanguine. He recently predicted that coding as a career is doomed. Huang said that now that artificial intelligence makes it possible to code in simple English, anyone can become a programmer.
And one more thing! We can now use WhatsApp channels! Follow us so you don’t miss any updates from the world of technology. To follow the HT Tech channel on WhatsApp, click here Join now!
Follow us on Google news ,Twitter , and Join Whatsapp Group of thelocalreport.in
