• D.C.
  • BXL
  • Lagos
  • Dubai
  • Beijing
  • SG
rotating globe
  • D.C.
  • BXL
  • Lagos
Semafor Logo
  • Dubai
  • Beijing
  • SG


In today’s edition, a look at “Prometheus,” Microsoft’s new AI model designed to be integrated with ͏‌  ͏‌  ͏‌  ͏‌  ͏‌  ͏‌ 
 
cloudy Redmond, Wash.
sunny San Francisco
cloudy Beijing
rotating globe
February 8, 2023
semafor

Technology

Technology
Sign up for our free newsletters
 
Reed Albergotti
Reed Albergotti

Hi, and welcome to Semafor Tech, a twice-weekly newsletter from Louise Matsakis and me that gives an inside look at the struggle for the future of the tech industry. I spent most of yesterday at Microsoft’s campus in Redmond, Washington for a mystery product announcement that turned out to be for “Prometheus,” an AI model designed to be integrated with Microsoft’s products.

Prometheus is built with next-generation artificial intelligence from OpenAI, the company behind the wildly popular ChatGPT bot (Semafor broke the news that this was coming last week). Read on for what this means for Microsoft and its rivals. I also have early access to Prometheus and will be testing it out this week, with an update in Friday’s newsletter.

Speaking of Prometheus, I asked a New York baby name consultant what she thinks about the monikers for the new chatbots. Plus, a TikTok-themed “Saturday Night Live” skit, as well as Louise’s thoughts on a subversive new AI project that could influence the future of copyright law.

Are you enjoying Semafor Tech? Help us spread the word!

Move Fast/Break Things

➚ MOVE FAST: Baidu. The search giant announced it’s almost done testing its own ChatGPT-style bot, Ernie, which is based on an artificial intelligence model the company first introduced in 2019. The move caused Baidu’s stock price to soar.

➘ BREAK THINGS: Other Chinese tech firms. E-commerce giants Alibaba and JD quickly tried to hop on Baidu’s coattails, saying they were working on similar technology. But the projects are less of a natural fit for the companies, and investors don’t seem convinced. While Alibaba’s stock slightly increased, it failed to get a big bump like Baidu’s did.

Reuters/Thomas Peter
PostEmail
Semafor Stat

Number of people in the U.S. who have signed up for Twitter subscriptions as of last month, including Twitter Blue, according to an internal document viewed by The Information. Blue gives users access to exclusive features for $8 a month ($11 on Apple devices) and has become Elon Musk’s signature initiative at the social network. But the figures suggest that it’s only contributing $28 million in yearly revenue — a tiny sliver of the more than $1.2 billion Musk needs to make in annual interest payments on the debt he took on to buy Twitter.

PostEmail
Reed Albergotti

Microsoft has the edge in AI race, with the help of ChatGPT

THE NEWS

Some of San Francisco’s tech reporters were planning on attending a happy hour at House of Shields on Monday night, but organizers had to cancel it because so many of the guests were headed to Seattle.

It’s been eight years since Microsoft captured journalists’ and consumers’ imagination like this (With the HoloLens). We all flew north to see the unveiling of Prometheus, Microsoft’s specialized version of the artificial intelligence that powers the wildly popular ChatGPT. Prometheus will be integrated into the software giant’s Bing search engine, its Edge browser, and eventually, its entire line of business products.

In an address to a crowd of about 100 journalists and industry analysts, Microsoft executives demonstrated how Prometheus could instantly plan an itinerary for a family vacation to Mexico City, or tell you how to substitute an egg in a recipe, all without ever having to visit a web site.

Microsoft expects its new browser and search engine to instantly compete with Google, which rolled out its own answer to Prometheus, named Bard. Bing, which has never been able to steal significant market share from Google search, will become more powerful with the Prometheus artificial intelligence.

“The race starts today,” Microsoft CEO Satya Nadella said.

REED’S VIEW

Tuesday’s news wasn’t really a product announcement. To me, it was a complete reorientation of Microsoft around this new breed of AI that has taken the world by storm, the same way Bill Gates pivoted to the internet in the 1990s.

In that sense, the event is broader than just Microsoft, which makes the bulk of its revenue selling services like Office products and cloud hosting to other companies. The entire technology industry is about to be upended.

Microsoft is uniquely positioned to benefit from the shift. Many of the applications of these AI “large language models,” which are capable of digesting seemingly infinite sources of data, will be more valuable to companies than they are to individuals. And serving companies is in Microsoft’s wheelhouse, more so than it is for Google.

Microsoft CEO Satya Nadella
Reuters/Jeffrey Dastin

It’s clear that this transition has been in the works for years and that Nadella saw this AI revolution coming and played the long game. By the time Microsoft first invested in ChatGPT creator OpenAI in 2019, it had already begun to build the institutional muscle necessary to capitalize on the breakthrough, investing in the safety guardrails for the technology as far back as 2017.

OpenAI brought Microsoft a key ingredient: Cutting edge AI research that broke through the previous limits on how much data these models could digest.

Microsoft has now spent years building highly specialized servers optimized to run OpenAI models for millions of users. And the number of users is really the key to making this technology practical.

Mind blowing AI product demonstrations are one thing. But if that technology can’t be run efficiently on servers, it isn’t going to make it to market. Microsoft’s advantage may not be the AI technology itself. It’s all the infrastructure needed to make it work in an economical way. Whether that takes weeks, months or years is the key question.

Microsoft’s competitors — led by Google — are now playing catch-up. On Wednesday, Google showed off its Bard technology at an event in Paris, but it was not the main focus of the event, which included other new product announcements.

Facebook is also a big player in the AI game and has released its own chatbots. But it’s unclear how it can profit from it in the same way that Microsoft and Google can.

And then there’s Apple and Amazon, which seem to be left behind in the AI arms race. This isn’t to say they can’t catch up. This revolution will take years to play out. Nobody really knows where it’s headed.

In fact, the very nature of this technology is not well understood by Microsoft, Google or the researchers who have pioneered the underlying models.

One example: Prometheus, which I got early access to and have been playing around with, seems to have multiple personalities. The same question won’t always result in the same answer the next time you open up a browser window and begin a new chat.

The reason for this isn’t completely understood, but it may have something to do with the way the work is divided up among the hundreds of thousands of central processing units searching for the answer. In essence, each time you open the window, you’re chatting with a new, unique entity.

A lot is unknown, but one thing is certain. Microsoft has already beaten its competitors in at least one area. It came up with the best name for its AI. Bard, Claude and Ernie just don’t compare to Prometheus, who stole fire from Zeus to give to humanity, which he created out of clay.

PostEmail
Evidence

PostEmail
Ahem

Reuters/Guglielmo Mangiapane/File Photo

For some people — let’s call them seasoned — Saturday Night Live offered a great Gen Z internet vocabulary lesson this past weekend in a skit starring Pedro Pascal, the lead in HBO’s new hit show The Last of Us. Even if you don’t know what “left no crumbs” means, don’t understand why Pascal is “daddy,” or have never wasted hours of your life watching “fancams” on TikTok, you will laugh at this sketch. (This should not be read as an admission of ignorance by the author.)

Befuddling the olds is a right of passage for every generation, but the sketch spoke to a deeper truth about the internet today, where pop culture has become divided into narrow interests. Another sketch starring Pascal touched on the same theme: It portrayed a film and television trivia game show in which all three contestants seemed to know every answer, so long as it referred to works that were not produced in this century.

The paradox of the explosion of online entertainment is that it has something for everyone, but is no longer the unifying cultural force that it once was. And instead of telling you what “left no crumbs” means, we’re inviting you to send us your best definitions at reed.albergotti@semafor.com. We’ll publish the most interesting ones in the next newsletter.

Reed

PostEmail
One Good Text

Sherri Suzanne, founder of baby naming consultancy My Name For Life.

We have been talking a lot about the human names tech companies give to their artificial intelligence tools. In the early aughts, many firms used feminine names or at least default feminine voices, like Alexa, Siri, and Cortana, and later faced backlash for assigning female traits to their automated “assistants.”

The latest batch of chatbots, however, have more masculine names in the mix, like Anthropic’s Claude, Google’s Bard, and Baidu’s Ernie. OpenAI, of course, has gone with a definitively robot-sounding name: ChatGPT. Reed asked Suzanne what she makes of the trend.

PostEmail
Watchdogs

In other Microsoft news, Britain’s antitrust regulator said today that the software giant’s $75 billion deal to acquire Activision could “result in higher prices, fewer choices, or less innovation for UK gamers.”

The country’s Competition and Markets Authority, which will decide whether to let the deal move forward by the end of April, said it would ask the companies to come up with ways to ease those concerns. It’s another big hurdle for Microsoft after the U.S. Federal Trade Commission sued to block the transaction in December.

PostEmail
Obsessions

arXiv

One of the key legal questions still to be determined about generative AI tools is whether their outputs violate copyright laws. Programs like OpenAI’s DALL·E and Stability AI’s Stable Diffusion were trained by scraping millions of photographs, collages, paintings, and illustrations made by humans from all across the internet. Getty Images, which is currently suing Stability AI in the U.S. for allegedly using its archive without permission, argues training AIs should not be considered fair use.

Two recent developments may help support Getty’s argument. First, a group of researchers from Google and several universities published a paper showing that Stable Diffusion sometimes spits back the images it was trained on, rather than new mashups. Second, the AI startup Chroma released a beta project called Stable Attribution, which its co-creator claims can find the “the images in the model’s training set which most contributed to the generated image.”

In other words, you can create an image using Stable Diffusion, and then use Stable Attribution to find the artists whose work most closely resembles what the AI generated. The tool is a fascinating turn in the debate over artificial intelligence and copyright, with the legal outcomes possibly stymying or encouraging one of the most-hyped technologies today.

Louise

PostEmail
How Are We Doing?

Are you enjoying Semafor Tech? The more people read us, the better we’ll get. So please share it with your family, friends and colleagues to get those network effects rolling.

And hey, we can’t inform you on what’s happening in tech from inside your spam folder. Be sure to add reed.albergotti@semafor.com (you can always reach me by replying to these emails) and lmatsakis@semafor.com to your contacts. In Gmail, drag this newsletter over to your ‘Primary’ tab.

Thanks for reading.

Want more Semafor? Explore all our newsletters at semafor.com/newsletters

— Reed and Louise

PostEmail