2023-08-18 07:56:00
« In the near future, nWe are going to have Copilots for everything, and we are no longer going to want to work without “, predicted with the Tribune Thomas Dohmke, CEO of GitHub and one of the pioneers of artificial intelligence-powered assistants. In October 2021, his company, a subsidiary of Microsoft, launched Copilot, a tool capable of suggesting to developers the continuation of their computer code. It was one of the first so-called “generative” AI applications, even before the term was commonly used.
More than a year later, in the wake of the resounding success of ChatGPT, Microsoft is rushing to integrate artificial intelligence-enhanced assistants into all its products, starting with the Microsoft 365 office suite (Excel, Word, PowerPoint. ..), its Teams communication software and its cybersecurity tools. Rather than multiplying the names of assistants, it borrowed from its subsidiary the Copilot brand (copilot, in French), which it declines for each software. And for good reason: this term, coined by a former GitHub employee, perfectly reflects the group’s strategy in generative AI, which might quickly prove to be very lucrative.
$30 per user per month
« The term copilot works well because it conveys the fact that the tool always needs a pilot, be it a developer, a writer, or a sales person. The message is clear: it does not replace the human worker. It is simply a probabilistic machine that helps us do our work faster by eliminating repetitive tasks. », develops Thomas Dohmke. Depending on the applications, Copilot suggests computer code, writing or images, but it is up to the user to sort and validate the information. For example, Microsoft 365 Copilot (still in testing with a small number of users) can create a complete PowerPoint presentation in seconds from information in Word documents, emails or Excel spreadsheets. It therefore takes on the burden of laborious and time-consuming office work, so that the user can concentrate on other aspects of his presentation.
With its Copilot, which will be rolled out to the general public in the coming months, Microsoft promises businesses unprecedented productivity gains. To convince the market, the computer giant can already rely on the first figures produced by its subsidiary Github thanks to its 20,000 client organizations. ” GitHub Copilot costs companies $19 per engineer per month, for productivity gains we estimate between 30% and 50%. It’s phenomenal, especially if you put it in relation to the hourly wages of users, who are employees highly qualified et so well paid “, thus promotes Thomas Dohmke.
« CThis revolution is timely for companies since in the current economic context, everyone wants to reduce costs “, he adds. Specifically, GitHub has claimed since February that around 46% of the computer code written by Copilot users has been suggested by AI, a share that rises to 60% for computer languages such as Java. Thanks to the assistant, the company has finally exceeded one billion dollars in annual turnover in 2022, and plans in the short term to become one of the financial pillars of its parent company.
For its part, Microsoft plans to charge its assistants even more expensive than GitHub, up to more than 30 dollars per user depending on the software, which would create a gigantic source of revenue to exploit. In the last quarter, the company said, for example, it sold 383 million paid Microsoft 365 accounts, or as many customers as it can hope to convince to pay an additional $360 per year. In other words, the tech giant is not content to use generative AI as a differentiating asset from the competition, it is also making it a new revenue lever in its own right thanks to assistants.
Towards a universal Copilot?
While the idea of using the Copilot brand for all assistants is above all a matter of marketing, it also makes it possible to anticipate potential developments. “ We are heading towards a future where Copilots will share a common layer of knowledge », Projects the CEO of Github. The Microsoft 365 Copilot is a first example of this vision, since it has access to information from all the different software in the office suite. But the future would go even further, and would be able to follow the user through different applications, starting with those of Microsoft.« Let’s imagine that I have a Teams meeting with one of my employees, he might ask Copilot to fetch in a few seconds on Github the passage of his project on which he is blocking to discuss it with me », Aims Thomas Dohmke.
However, this evolution towards more powerful assistants capable of performing more and more tasks will have to be accompanied by technical adjustments. The AI models that serve as their engine become more and more precise in their production with each iteration (like GPT-4), but they require in return more computing power, and therefore time.
To meet this need, the semiconductor industry, starting with Nvidia, is developing specialized chips. Not only are the power of these new processors able to run the latest AI models, but the older models are also becoming faster to run, and therefore cheaper to use. ” If processors progress faster than features, then the question of cost will not arise. », sweeps Thomas Dohmke. And to add: the architectures of long-released models continue to be optimized, making them cheaper to run. In concrete terms, the latest version of GitHub’s Copilot uses different AI models depending on its tasks: for those where precision prevails, it uses the most recent versions of GPT, and for those where execution speed is essential, it will use older versions.
What regarding the consequences on employment?
Problem: if Microsoft insists on the place of humans at the helm of AI and its Copilot assistants, it remains at the same time very discreet regarding the potential harmful effects of these new tools on the labor market. However, a big question already appears in the background: what are companies going to do with all the promised productivity gains? Among the many possible consequences of increased worker efficiency is of course the specter of job closures. And the example of GitHub did not make it possible to rule out this threat, since its Copilot is aimed at trades under pressure, including that of developer.
But the suspense around this question should not last very long, because the revolution is well and truly underway. ” Before, our potential customers did not feel any urgency to adopt generative artificial intelligence, even if they agreed that it had a future. They didn’t want to be the first. ChatGPT brought that sense of urgency concludes Thomas Dohmke.
1692489337
#Copilot #assistants #Microsofts #gold
