On Monday, Microsoft launched an OpenAI service as part of its Azure cloud platform, offering businesses and start-ups the ability to incorporate models like ChatGPT into their own systems. The company has already been building AI tools into many of its consumer products, such as a DALL-E 2 feature in its Bing search engine that can create images based on a text prompt, and the Information reported recently that it’s working to bring more of them to Microsoft Office as well.
Eventually, CEO Satya Nadella said in Davos this week, “Every product of Microsoft will have some of the same AI capabilities.”
The company is staking its claim to what many in the tech industry believe will be an AI revolution. Microsoft is one of several, including rival Google, betting that this generation of artificial intelligence will transform not just productivity software but entire industries, thanks to so-called large language models that can understand, converse with, and imitate humans in realms ranging from writing to art to computer coding.
“I think this is going to radically change the future for all knowledge work,” said Bojan Tunguz, a machine learning modeler and data scientist at Nvidia, of the recent advances in large language models. “We’re only seeing the early days of what that entails.”
That’s the bullish view, and it’s one that you’ll hear a lot in Silicon Valley and Seattle these days. While Google, Microsoft, and other tech giants have been testing and honing AI models for years, a new crop of risk-taking upstarts has stolen their thunder by taking their experimental AI programs directly to the public. The release of tools like ChatGPT, DALLE-2, Stability AI’s Stable Diffusion and Midjourney in the past year has made “generative AI” — software that draws on vast data sets to create novel works — the industry’s buzziest byword.
Now that they’ve caught on, Big Tech is playing catch-up. And no one’s playing it harder than Microsoft.
At Davos, Nadella predicted that the current generation of AI will spark an industry-wide “platform shift” on par with the moves to mobile devices and cloud computing over the past 15 years. Microsoft, which was an early investor in OpenAI, is reportedly planning to take a much larger ownership stake in the company in what could be a $10 billion deal, the news site Semafor first reported last week. Microsoft and OpenAI declined to comment.
There’s also a skeptical view, in which the technology proves dazzling as a toy and a novelty but underwhelming or even harmful in its practical applications.
ChatGPT can produce remarkably plausible-sounding text on a wide range of topics, but it’s prone to factual errors and problematic biases, and some school districts have already banned it as a potential cheating tool. The tech news site CNET is under fire for quietly using AI to write articles, some of which were found to contain errors. Stability AI, the maker of AI art generator Stable Diffusion, has been sued by Getty Images for allegedly training its model on copyrighted works without permission. Microsoft has its own history of AI missteps, including the 2016 release of a chatbot called Tay that trolls trained to embrace genocidal hatred.
Those risks help to explain why Google, which has developed some of the most advanced AI chat tools, has yet to release them to the public. Google’s LaMDA chatbot system is so sophisticated that one of the company’s engineers became convinced it was sentient, reigniting a debate over whether it would be irresponsible to make similar tools available to ordinary users.
“Some people are going to get hurt — that’s inevitable,” said Meta’s chief AI scientist, Yann LeCun, on Thursday at a forum hosted by the company Collective[i]. That shouldn’t stop the march of progress, he added, but it is important for companies involved in developing new forms of AI to find ways to mitigate the damage.
Downplaying ChatGPT as “just nicely done, not revolutionary,” LeCun suggested that the reason it came from a start-up rather than a tech giant is because “Google and Meta both have a lot to lose by putting out a system that makes stuff up.” (In November, Meta released a language model for scientists, called Galactica, only to pull the plug on it three days later due to a backlash.)
Still, Microsoft’s embrace of OpenAI puts renewed pressure on its rivals, particularly in the lucrative cloud computing sector. While using ChatGPT to improve its own products could help Microsoft hold its edge in productivity software, the larger battle is to sell AI services to businesses. Those could include established firms looking to build a smarter customer-service chatbot, start-ups developing more specialized AI tools, or even other AI companies that need cloud computing power to train their own models.
One application that’s already gaining traction is the use of AI to assist software developers in writing code. Microsoft’s subsidiary GitHub uses OpenAI technology in a tool called GitHub Copilot, which can suggest code in real time as you’re programming.
“Definitely Microsoft adopting it will make it a lot more widely available,” said Sridhar Ramaswamy, CEO of the start-up Neeva, which recently introduced an AI-based search engine that uses language models to answer users’ questions directly. “The problem is, once Microsoft starts turning the screws on a particular area, they tend to squeeze out everybody else.”
Ramaswamy, a former Google executive, predicted that Google will release its own chatbot in the next month or so, “and it’ll be very good.” But in the long run, he said he doubts Microsoft or Google will be able to corner the market.
“The tech is rapidly getting commoditized,” Ramaswamy said, with start-ups such as Anthropic and Cohere already building their own tools and models. “Start-ups can afford to take risks, because they have a lot less to lose. I think that’s going to be an issue with Google and Microsoft.”
On Friday, Google’s parent company cut 12,000 jobs, as CEO Sundar Pichai told the remaining employees that the company would sharpen its focus on AI. Earlier this week, Google AI guru Jeff Dean published a blog post highlighting the company’s latest work on large language models, and the New York Times reported that the company has called on its founders, Larry Page and Sergey Brin, to review its AI strategy.
Meanwhile, Microsoft’s big moves in AI lend the 47-year-old company an air of excitement at a time when gloom pervades the tech sector. Microsoft announced its own cuts on Wednesday, following rounds of layoffs by Amazon, Meta and others. But its aggressive AI investments could give its remaining employees — and future recruits — a reason for optimism.
To deliver on that promise, it will have to show that it can turn its shiny new AI tools into something more consequential than Clippy 2.0.