Microsoft kickstarts the AI arms race
The relaunched Bing does research, writes emails, and just might hit Google where it hurts
REDMOND, WA — In the end, it might not matter who moved first. The first round of search engine wars saw several entrants rise and fall before one emerged as the victor, and Google undoubtedly benefited from watching all the mistakes that Yahoo, Infoseek and all the rest made along the way.
At the same time, as I interviewed executives and watched presentations about the next generation of Bing on Tuesday, I was struck by the once-in-a-generation opportunity Microsoft may now have to shift consumer behavior. In December, running my first few queries on ChatGPT, I argued that the technology almost certainly represents the future of search.
Today, barely two months later, the future arrived in a browser on my desktop computer.
"It's a new day in search,” Satya Nadella told reporters as the day begun. “It's a new paradigm for search. Rapid innovation is going to come. The race starts today."
There was an exultant quality to Microsoft executives’ remarks on Tuesday, as they walked through how the company’s partnership with the insurgent San Francisco firm OpenAI would allow it to offer an improved version of its hit demo ChatGPT at scale.
Using the reimagined Bing on stage, a top marketing executive at the company planned a five-day trip to Mexico City, identified the best 65” television, and then refined his query to find the best 65” TV for gaming. He generated a list of history’s top Japanese poets, augmented with sample haikus, and then put together a quiz about music in the 1990s.
All of this he did more or less instantly, with the fluidity that is now familiar to users of ChatGPT, but bolstered by the knowledge that soon this tool would be available to the masses, and for free. (Microsoft said there will be unspecified limits on the number of queries users can run, at least at launch, but for the moment does not have plans to charge a subscription the way OpenAI has.)
Demonstrations like these often inspire hyperbole on the part of tech company presenters, but Microsoft executives seemed sincerely wowed by what they had managed to build. Sarah Bird, the company’s responsible AI lead, said of the new Bing: “It's the most exciting and powerful technology I have ever touched.”
I finally touched the new Bing at 2PM PT, when Microsoft enabled the feature on a test build of Edge I downloaded to my laptop. A new Bing button in the top right-hand corner of the browser opens a sidebar with two of the company’s core new features. The Chat tab allows you to submit queries of up to 2,000 characters, with Bing remembering the context of your previous questions as you go.
One tab over, Compose generates text for you based on your inputs. It currently offers five “tones” — “professional,” “casual,” “enthusiastic,” “informational,” or “funny — and lets you choose whether to generate output as a paragraph, an email, a blog post, or a list of ideas. (You can also choose whether the output is short, medium, or long.) Tap “generate a draft” and you can then copy the text elsewhere for editing.
I expect both of these features will eventually find hundreds of millions or even billions of users, just as soon as Microsoft will allow them to try it. (ChatGPT hit an estimated 100 million users in January after just two months of existence.)
The “when” factor here is unusually interesting. Google, irritated about all the good press ChatGPT has been getting, tried to front-run Microsoft by pre-announcing its own conversational AI tool Bard on Monday. It will hold another AI event in Paris on Wednesday. Yesterday, Google CEO Sundar Pichai CEO said Bard would be available “in the coming weeks.”
If we expect that Bing and Bard will be broadly comparable, as screenshots of the latter product suggest they will be, then Microsoft’s first-mover advantage here might not last through February.
But the unprecedented growth of ChatGPT suggests that even a few weeks might give Microsoft an enduring head start — or at least enough runway to begin chipping away at Google’s search dominance. Peeling off even 1 percent of Google’s market share would be worth $2 billion a year in advertising revenue to the company, executives told investors on a call Tuesday afternoon.
Of course, there are plenty of serious questions about the broader implications of these tools becoming widely available. Microsoft spent a good deal of time Tuesday explaining its approach to making the new Bing safe, which largely involved describing org charts accompanied by arcane diagrams. Ultimately nobody knows how safe it is to put AI tools like these into the hands of billions; to some degree everyone here is just hoping for the best.
There are also the regulatory concerns, which came up less often but seem just as relevant to me. I’ve written before that publishers seem likely to sue over the way that their journalism was scraped to train the models on which tools like these run; I wouldn’t be surprised if at least four European countries started drafting lawsuits against Bing today based solely on what they could see in Microsoft’s screenshots.
So how good are the new tools?
As I write this, I’ve had less than two hours to put the technology through its paces. So far, it seems broadly similar to what ChatGPT can do. My big hope was that Bing would seem obviously superior because, unlike its predecessor, its model is frequently updated with new information. (The training data for ChatGPT ends in 2021.)
In some of these cases, Bing really shines. I asked both Bing and GPT: “Why did Russia start a war in Ukraine?” ChatGPT responded with one paragraph about the annexation of Crimea in 2014. The new Bing, on the other hand, told me that “The reasons for Russia’s invasion of Ukraine are complex and disputed,” but offered five possible causes: each one linked to an original information source with a footnote. It also linked me to the relevant Wikipedia page. The user experience there is quite good.
Bing can also have strange misses, though. When I asked it what I thought was a simple question — when did Elon Musk take over as Twitter CEO? — it threw up its hands, possibly because people have told too many jokes about the subject.
“I’m sorry, but I cannot answer your question based on the web search results,” the bot told me. “The search results are inconsistent and unreliable, and some of them are from parody or satire sources. Please be careful when reading the search results and verify the information from credible sources.”
On one hand, this is a disappointment: these tools truly won’t be living up to their promise until they can answer basic fact-checking questions like the Musk one. On the other hand, I was pleased to see the bot owning up to its limitations, rather than attempt to hallucinate its way through an answer.
I also tried to compare Bing and Bard, using the extremely limited data set I had available. Google’s screenshot of Bard features the query “what new discoveries from the James Webb Space Telescope can I tell my 9 year old about?“ I compared the answers Google showed to the same query on Bing.
Google offered three examples of the telescope’s discoveries in its response, according to the screenshot; Bing offered me six. Unlike Google, Bing also linked to its information sources. (Maybe the production version of Bard will cite its sources, too. Here’s hoping!)
After the keynote presentation, my Hard Fork co-host Kevin Roose and I sat down with Microsoft CTO Kevin Scott and OpenAI CEO Sam Altman. (We’ll share more of that conversation on Friday’s episode.)
Altman told us that AI “will be as big of a deal as any of the great technological revolutions that have come before it, in terms of what it means for enabling human potential.”
“I think Kevin and I both very deeply believe that if you give people better tools, if you make them more creative, if you help them think better, faster, be able to do more …people will change the world in unbelievably positive ways,” Altman said.
Whether we come to view AI as an unqualified good remains up in the air. But after the events of this week, it no longer feels like hyperbole to suggest that AI really is going to change things — starting with search and publishing, extending next into productivity tools, and continuing to march onward from there.
The race, as Nadella said, starts today. And its winners — and losers — remain very much yet to be decided.
Elsewhere in AI: Microsoft plans to let companies soon build their own custom versions of ChatGPT. And here’s how the intense criticism Meta faced over misinformation and harmful content has made it hesitant to roll out its AI products more broadly, despite making massive investments.
Talk about this edition with us in Discord: This link will get you in for the next week.
Governing
The Department of Justice wants a jury to decide whether Google should have to break up its advertising business, a departure for antitrust cases, which are usually decided by a judge. (Rana Foroohar / Financial Times)
Twitter is expected to fall under the European Union’s new Digital Services Act, meaning it has enough EU users that it will have to report to regulators on how it’s reducing harmful content. (Jillian Deutsch and Kurt Wagner / Bloomberg)
India blocked TikTok in 2020 and users turned to Instagram and YouTube instead, blunting the impact of the ban on creators. (Russell Brandom and Nilesh Christopher / Rest of World)
Industry
Children ages 4 to 18 now spend an average of 107 minutes per day, or 60 percent longer than the time they spend watching videos on YouTube. Going to go out on a limb and suggest that’s more time than your 4-year-old should ideally be spending on TikTok! (Sarah Perez / TechCrunch)
Meta is asking more of its middle managers to become individual contributors or leave the company in an effort to flatten its hierarchy. (Sarah Frier and Kurt Wagner / Bloomberg)
Meta is revamping its fledgling Horizon Worlds app and opening it up to teens ages 13 to 17. (Salvador Rodriguez / Wall Street Journal)
Mastodon’s growth didn’t last — the decentralized social network now has nearly half a million fewer total registered users than it had at the start of the year. It’s still more fun than Twitter though! (Amanda Hoover / Wired)
While some school districts are banning ChatGPT, other teachers are asking students to think critically about the technology, and incorporating it into lessons. (Natasha Singer / New York Times)
Google will now automatically blur explicit images by default in search results. (Jess Weatherbed / The Verge)
Pinterest shares slipped in after hours trading after the company reported revenue that missed analyst expectations and issued a light forecast for the first quarter. (Jonathan Vanian / CNBC)
Tinder rolled out new safety features including incognito mode, which allows users to stay hidden while scrolling through peoples’ profiles. (Aisha Malik / TechCrunch)
Those good tweets
For more good tweets every day, follow Casey’s Instagram stories.
Talk to us
Send us tips, comments, questions, and Bing searches: casey@platformer.news and zoe@platformer.news.