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


In today’s edition, an interview with a Microsoft executive on new AI-powered products, Sam Altman o͏‌  ͏‌  ͏‌  ͏‌  ͏‌  ͏‌ 
 
rotating globe
January 17, 2024
semafor

Technology

Technology
Sign up for our free newsletters
 
Reed Albergotti
Reed Albergotti

Hi, and welcome back to Semafor Tech.

For the vast majority of people, generative AI hasn’t changed their life much. That probably won’t happen until it starts upending the software they already use every day: email, calendar, word processors, spreadsheets.

Last summer, I went to a product launch in New York where Microsoft demoed some of the generative AI capabilities woven into its operating system and Office suite of products. It seemed like a threshold had been crossed in the AI revolution — the point at which chatbots go from novelties to time-saving game changers.

There was just one frustrating thing: I couldn’t use it. As you’ll read below, Microsoft slowed down the rollout of these products for several reasons, limiting them to enterprise clients and early testers that did not include reporters (who almost all use Google’s version of Office).

Earlier this week, Microsoft finally made it public. This, along with OpenAI’s GPT store, could lead to the kinds of productivity gains that we’ve been expecting from this technology. I spoke with Yusuf Mehdi, Microsoft’s consumer chief marketing officer, about the rollout. Read below for the edited interview.

Plus, I’m in Davos this week and AI is THE hot topic. We talked to a tech veteran about the hype and take you inside the first joint public appearance of Sam Altman and Satya Nadella since OpenAI’s meltdown. If that’s not enough for you, sign up for our Davos pop-up newsletter.

Move Fast/Break Things
Reuters/Loren Elliott

➚ MOVE FAST: Apple. Even with lagging iPhone sales, the company topped Samsung for the first time in global smartphone shipments in 2023, according to IDC. The last time the Korean firm came in second was in 2010, when Nokia was No. 1. The overall market is also expected to recover this year, which is more good news for Tim Cook’s company.

➘ BREAK THINGS: Apple. The U.S. Supreme Court declined to come to its aid in its fight with Epic Games over the App store. That could allow developers to point users to buy things outside of Apple’s in-app payment system, and the iPhone maker revised its policy in response. As a result, legal battles may continue while Apple faces similar pressures in Europe.

PostEmail
Q&A

Yusuf Mehdi is Microsoft’s executive vice president and consumer chief marketing officer.

Q: Why did it take so long to roll this out to everyone? Were there technical hurdles, or scale hurdles on the server side?

A: All of the above. We wanted to start focusing clearly on where we could add value. We wanted to prove the technology out, make sure that we understood the highest use case for enterprise customers. These are obviously costly services to use, too, because of the GPUs. We’ve learned a lot about how to do that, too. We’re at a point now where we can scale more. We’ve added more GPUs. We’re ready to service more customers. So all of that has come to bear to allow us to get to this point.

Q: And the cost is $20 a month for consumers and $33 per user, per month for enterprise customers. At that price, how much room do you have in the margins and what kind of tricks do you have to manage the costs?

A: We’ve priced it out in such a way that we feel good about the fact that we’re scaling now to a broader audience. And we’re feeling great about the Copilot Pro and what we think we can accomplish with that rollout, too.

There’s a lot of stuff we’re doing to reduce costs. Everything from different technology optimizations, new approaches to how we run things on the backend server. There’s a wide variety of things we do to continue to reduce the cost so that we can bring that value to more people and give them more benefits.

Q: Is there anything that’s not generally available?

A: There’ll be things that will fast follow, like GPT Builder. So the ability to build GPTs will come very quickly.

Q: Tell me about GPT Builder.

A: Our intention is to make these interoperable with OpenAI. You’ll be able to use them from within the OpenAI app or call on OpenAI GPTs created with the OpenAI app.

Microsoft

Q: How much is this product going to get to know you and learn your habits? Will I know if I missed an important email?

A: You will. We’ve been working on doing things more in your voice. So it’ll understand how you write. Do you use punctuation? Do you put in exclamation points? Getting the email to understand your voice and write more in your voice. And then your chat history is a way to start remembering things you’ve done. This is ripe for personalization to get more out of it.

Q: How do you see the competition with Amazon, which is offering enterprise products that work on all Microsoft products. How interoperable will you be across products that compete with Microsoft ones?

A: You should think of us as also having very similar platform agnosticism. We want to have our technology run everywhere. That is the plan. That’s why Office runs on Mac, runs on Android, runs everywhere. We’re going to do the same thing here.

The Copilot app runs on iOS, it runs on Android, it’ll run anywhere you can get access to it. So that’s a value prop today. And then we will build more of that capability. The differentiation we have is the apps that people use to create and be productive, plus AI. The work graph, the knowledge graph. You can say, ‘who said this in a meeting?’

Or it’s hard for me to keep track of all the new terms. This came up with retrieval-augmented generation. In a meeting, they are using the RAG acronym. I can ask, ‘what is RAG?’ and it comes back that this has been referenced in these three documents that were reviewed with the senior leadership team. Here’s the PowerPoint, you can go right here. But the summary is that it’s a new theory about how to get more relevant AI answers.

I was in a meeting and I did that five times. That was just my ability to catch up on things that I had missed, and then get right to the point of the document. All that happens because we have that data graph.

Q: What’s your favorite thing to do with this?

A: I use it for so many things. My son hurt his shoulder and he has to get surgery on it. So he got an MRI. And we were trying to make heads or tails of the MRI. It was very hard. So I took that whole report and put it in and said, ‘tell me in plain English what this is saying.’ It came back and said, ‘There are five things apparently potentially up with your son’s shoulder. He’s got a torn labrum, he’s got a chip in the bone.’ And then it had a great paragraph for each thing. We went from not understanding anything to ‘Oh, wow, I have a sense of it.’

PostEmail
One Good Text

Neil Lawrence is the inaugural DeepMind Professor of Machine Learning at the University of Cambridge.

PostEmail
Obsessions

The hot panel at Davos today, at least for the tech crowd, was a discussion with OpenAI CEO Sam Altman and Microsoft CEO Satya Nadella. The last time I saw these two on a stage together was at OpenAI’s developer day, just before Altman was fired by OpenAI’s board, starting one of the greatest and strangest corporate dramas in recent memory. Nadella played a central role in that drama, at first offering to hire every employee at OpenAI and then aiding in a peace deal.

But in a packed, stuffy room inside Microsoft’s Davos headquarters, the only people not sweating seemed to be Altman and Nadella. The drama was only mentioned in passing. The message, instead, was that AI is kind of boring. “I believe that someday, we will make something that qualifies as an AGI by whatever fuzzy definition we want,” Altman said. “The world will have a two-week freak out. And then people will go on with their lives.”

He added that he thinks the capability of GPT-4, the most advanced language model in the world, is “bad” but that gradually, over time, the company will build something that’s good. If, for some reason, OpenAI builds a language model that is too powerful or too dangerous, he says the company will decide not to release it to the masses.

The Economist/X

I’ve been thinking about the adage, “artificial intelligence is anything a computer cannot yet do,” as I bounce around Davos this week, meeting with people in the field. There was a time around 2014 when every tech company suddenly became an “AI” company. What they were really doing was data science, using machine learning and occasionally deep learning. But we eventually stopped calling that stuff AI.

When ChatGPT came out, it seems we started referring to those techniques as AI again, as if they were swept up by a giant LLM hype wave. The reality is, the term “AI” really has little meaning. A better term is probably simply “automation,” which humans have been working to perfect for thousands of years. Automation doesn’t sound as exciting, but it’s also not as scary.

PostEmail
Hot On Semafor
PostEmail