Microsoft AI leaps forward. Here’s how its human leader thinks about it | Tech Rasta


Earlier this year, LinkedIn co-founder and venture capitalist Reid Hoffman issued a startling warning about AI. “There is literally magic happening,” Hoffman said, speaking to technology executives across sectors of the economy.

That magic is even more evident in creative spaces like the visual arts, and the idea of ​​”product technology” has caught Silicon Valley’s attention. AI has recently won awards at art exhibitions.

But Hoffman’s message was aimed at executives.

“AI will change all industries,” Hoffman told members of the CNBC Technology Executive Council. “So everyone should be thinking about that, not just in data science.”

Copilot AI, an automated code writing tool from the GitHub open source subsidiary, is rapidly developing. Microsoft, Hoffman on Microsoft’s board is one example, cited directly as a sign that all organizations should be better prepared for AI in their world. Even without investing heavily in AI today, business leaders must understand the pace of development and upcoming applications in artificial intelligence or they will “sacrifice the future,” he said.

“Over 100,000 developers took 35% of their coding instruction from Copilot,” Hoffman said. “That’s a 35% increase in productivity, and that’s not in last year’s model. … In everything we’re doing, we’ll have the tools to expand, which will get there over the next three to 10 years, a baseline for everything we’re doing,” he added.

Copilot has already added another 5% to Hoffman’s stated 35%. GitHub CEO Thomas Domke recently told us that Copilot now handles 40% of the coding among programmers using AI during its beta testing period over the past year. In other words, for every 100 lines of code, 40 are written by AI, reducing overall project time by up to 55%.

A copilot, trained on large amounts of open source code, oversees the code the developer writes and acts as an assistant, taking input from the developer and making suggestions about the next line of code, often multi-line coding instructions, often requiring “boilerplate” code that would otherwise waste time for a human to recreate. We all have some experience with this form of AI now in places like our email, with both Microsoft and Google’s mail programs suggesting the next few words we want to type.

AI can reason about what might come next in a text string. But Dohmke said, “It can’t do much, it doesn’t capture the meaning of what you’re trying to say.”

Whether a company is a supermarket working on checkout technology or a banking company working on customer experience in an app, they are all effectively becoming software companies, all building software, and looking at developer productivity after having developers in the C-suite. And how to continuously improve it.

Here comes the 40 lines of code. “After a year of copilot, about 40% of the code was written by the copilot-enabled AI,” Domke said. “And if you show that number to executives, it’s mind-blowing to them. … Calculating how much they’re spending on developers.”

The logical conclusion is that with projects completed in less than half the time, there is less work for humans to do. But another way of looking at the job of a software developer is that they do much more valuable work than rewriting code that already exists in the world, Dohmke said. “The definition of ‘high value’ work is to remove boiler-plate menial work writing tasks over and over again,” he said.

Copilot’s goal is to help developers “stay in the flow” while they’re coding. That’s because some of the time spent writing code is actually looking for existing code from browsers to plug in, “snippets from somebody else,” Domke said. And that leads to distraction for coders. “Eventually they get back into editor mode and copy and paste the solution, but remember what they’re working on,” he said. “It’s like a surfer on a wave in the water and they have to find the next wave. The copilot is putting them in an editing environment, a creative environment and suggesting ideas,” says Dohmke. “And if the idea doesn’t work, you can reject it or find something close and always modify it,” he added.

GitHub’s CEO hopes to take more of those CoPilot code suggestions — up to 80% over the next five years. Unlike a lot in the computer field, Dohmke said of that forecast, “It’s not an exact science … but we think it’s going to grow tremendously.”

After being on the market for a year, the new models are said to be improving rapidly. As developers reject some code instructions from Copilot, the AI ​​learns. And as more developers adopt Copilot, a new colleague will become smarter by interacting with similar developers, learning from what’s accepted or rejected. New models of AI don’t come out every day, but every time a new model becomes available, “we might have a leap,” he said.

But AI still falls far short of replacing humans. “Today a copilot can’t do 100% of the job,” says Domke. “It’s not sentiment. It doesn’t create itself without user input.”

With Copilot still in private beta testing among individual developers — 400,000 developers signed up in the first months the AI ​​was available and hundreds of thousands more since then — GitHub hasn’t announced any enterprise clients, but hopes to start naming business. Users before the end of the year. No enterprise pricing information has been disclosed yet, but CoPilot pricing in the beta test is set at a flat rate per developer — $10 per month per person or $100 per year, often spent by developers on company cards. “And you can guess what they earn per month, so that’s the marginal cost,” Dohmke said. “If you look at 40% and think about productivity improvement and take 40% of the opex cost for developers, $10 is not a relevant cost. … I have 1,000 developers and it’s more than 1000 x 10 money,” he said.

The GitHub CEO sees what’s happening now with AI as the next logical step of productivity in the coding world he’s been a part of since the late 1980s. It was a time when coding was emerging from the punch card stage, and there was no Internet, and coders like Domke had to buy books and magazines and join computer clubs to get information. “I had to wait to meet someone to ask questions,” he recalls.

This was the first stage of developer productivity, then came the internet and now open source, allowing developers to find other developers on the internet who have already “developed the wheel”.

Now, whether the coding task is related to payment processing or social media login, most companies – startups or established enterprises – are put on open source code. “There’s already a huge dependency tree of open source,” Domke said.

It is not uncommon for up to 90% of the code in mobile phone apps to be pulled from the Internet and open source platforms like GitHub. In the age of “off-the-shelf” coding, that separates the developer or the app.

“AI is the third wave of this,” Domke said. “From punch cards to open source to building everything ourselves, so far in most code, AI is writing more,” he said. “With 40%, if AI spreads across industries, the innovation on the phone will be created with AI and developer help.”

Today, and into the future, Copilot remains a technology trained on code and making recommendations based on looking at things in a library of code. It’s not inventing any new algorithms, but with current progress, eventually, “with developer help it’s entirely possible to create new ideas of source code,” Domke said.

But it also needs a human touch. “Copilot is getting closer, but developers are always needed to create innovation,” he said.


Source link