Notes on Silicon Valley Girl: CEO of Microsoft AI: The Next 10 Years Will Change Humanity Forever, where Marina Mogilko talks to Mustafa Suleyman (CEO of Microsoft AI, DeepMind co-founder, author of The Coming Wave).
The 30-Second Reality Check
Mustafa Suleyman says hallucinations are “pretty much” solved, 20 minutes after saying you can’t rely on Copilot’s health answers yet. He also calls the desktop a billboard of ads. He runs AI at Microsoft.
Explain With Pictures
What Suleyman wants to replace Who put it there
┌──────────────────────────────┐
│ "all these windows with │
│ different colors and brands │ ◀── the company that shipped
│ ... a billboard of all │ Candy Crush in the
│ these competing apps" │ Windows 10 Start menu
└──────────────┬───────────────┘
│
▼
┌──────────────────────────────┐
│ your "humanist super- │
│ intelligence" (Copilot) │ ◀── also that company
│ connectors: Gmail, Dropbox, │
│ calendar, memory, voice │
└──────────────┬───────────────┘
│
▼
one assistant that sees everything,
"I don't mind" (as long as it's useful)The fix for one company owning your screen is, apparently, one company owning your context.
3 Hot Takes & Quotes
1. The desktop is a billboard. Yes. Who built it?
This is the part where I laughed out loud:
when you open your computer, you kind of have a billboard of all these different competing apps that are like, “Use me, try me”… it’s kind of distracting. It’s kind of ugly.
Correct! And Windows is where a lot of us first met that billboard. Candy Crush preinstalled in the Start menu, “suggested” apps, the tips that are ads. His answer is that the operating system, the browser, the search engine and the apps “are gonna slip away” behind your personal AI. Which is a very Microsoft way of solving the ads-in-the-Start-menu problem. You don’t remove the billboard. You become the only thing on it.
2. Hallucinations are “pretty much” eliminated. Also, please go see a doctor.
Around 17 minutes in, on Copilot’s health answers (which he says are about 40% of queries each week):
sometimes it makes mistakes, it’s not, you can’t rely on it yet. You always have to go see a doctor
Around 38 minutes in, on what he predicted in his book:
that was about hallucinations and I predicted that they were going to be largely eliminated and I think pretty much we’re on track. Pretty much they are.
Both can’t be fully true. Anyone who has used a coding assistant has seen it produce a neatly formatted config snippet with a flag that doesn’t exist, sitting among real flags and looking just like them. That’s the problem with “pretty much.” The 5% that’s wrong is perfectly confident and wears the same font as the 95% that’s right, and in a health chat nobody runs a unit test.
3. “It is going to remember everything… and be perfect.” And he doesn’t mind.
The memory pitch, in his words:
it is going to remember everything in the future and be perfect if you choose for it to do that.
Then, asked whether it worries him that AI knows so much about us:
No. I think that as long as it’s useful and it’s got a purpose, I don’t mind.
His own setup is the iPhone power button wired straight to Copilot voice mode, journaling about his day on the drive home, with the assistant keeping “state” for him like “a second memory.” That sounds genuinely useful. It also sounds like the most complete personal dataset anyone has ever handed one vendor. Plug in the connectors he describes (Gmail, calendar, Teams, Dropbox) and it gets more complete. Microsoft has been here recently, with Recall, the Windows feature that screenshotted your screen every few seconds and got pulled back in 2024 after security researchers found it all sitting in an unencrypted database. “I don’t mind” is a fine personal preference. It’s a weak security model.
The Post-Mortem / Verdict
The first answer of the interview sets the tone. Marina asks if we’re in an AI bubble:
Yes. So no, we’re not. I think so. I don’t think so.
After that he’s very bullish, and certain about things nobody is certain about, such as “I think we can prove that it doesn’t have subjective experience.” And then oddly careful elsewhere.
To be fair, the candid bits are real and he deserves credit for them. He says plainly that “these are labor replacing technologies” and expects “significant structural disemployment” by 2050, and he backs a version of Bill Gates’s robot tax, taxing capital because intelligence “is now getting converted into capital.” That’s further than most AI execs go on camera. His parenting advice, that kids still need “discipline and friction” to learn to teach themselves, is right too. Slightly awkward next to “probably not” on whether to save for your kids’ college, but right.
The episode ends on Copilot Groups, up to 32 people in one chat, where your AI doctor will talk to you for half an hour, answer 25 questions and reassure you before the human doctor joins for “a shorter time.” That’s the hospital’s dream, maybe. Patients may feel differently about it.
Links & References
- CEO of Microsoft AI: The Next 10 Years Will Change Humanity Forever | Mustafa Suleyman, Silicon Valley Girl Podcast, 14 Nov 2025
- Mustafa Suleyman (guest), CEO of Microsoft AI and DeepMind co-founder: X · Instagram
- Marina Mogilko (host), Silicon Valley Girl: Website · YouTube · Instagram
- The Coming Wave, Suleyman’s book
- Microsoft Copilot
- Microsoft Copilot introduces a group chat feature that lets up to 32 people collaborate in real time, The Decoder, on Copilot Groups