Meta’s Race for Your Face + Google’s Hit A.I. Notebook + HatGPT

1h 12m
Building great AR glasses is the hardest problem in consumer technology.

Listen and follow along

Transcript

At the University of Arizona, we believe that everyone is born with wonder.

That thing that says, I will not accept this world that is.

While it drives us to create what could be,

that world can't wait to see what you'll do.

Where will your wonder take you?

And what will it make you?

The University of Arizona.

Wonder makes you.

Start your journey at wonder.arizona.edu.

Now, Kisa, I had a very weird thing happen to me last weekend, which is that Netflix released a documentary that I was in.

So, this was something that I taped like two years ago, had basically no memory of, but it was with Bill Gates, and it's about AI.

I remember when you told me that you were going to do this, and it seemed like it was an exciting thing because it was going to be a documentary, and they were going to put it on Netflix.

Yeah, and then all of a sudden, I started getting texts from people I know with screenshots of their Netflix, where my face was the first thing that they saw when they logged into Netflix, like the big hero image on top of the thing.

Right, but except it wasn't really your face, was it?

No, it was my AI face because in the process of taping this thing two years ago, I made like a joke, what I thought was a joke, to the producers, but they were sort of like asking me, do you want hair and makeup?

And I said, no, because you can just use AI to make...

make me hot.

And they kept that in.

And then they actually did make me hot using AI.

So there's like a portion of the documentary where it just cuts to an AI version of me, but I am hot.

And the lower third graphic says hot Kevin Roos.

And that is what my friends and family members were greeted with when they opened Netflix.

You know what this reminds me of?

What's that?

In my 20s, I had a group of friends, and there was this other guy named Casey who kind of came into the friend group.

And they started calling him hot Casey because he was super, super good

And I was kind of looking around, like, guy, like, do we have to call this guy hot Casey?

Yeah.

And what one of my friends said to me was, like, you know, it could be worse.

They could be calling him okay-looking Casey.

That's true.

I guess Netflix could have titled that character like marginally attractive

Kevin.

Barely passable Kevin Roos.

I'm Kevin Roos, a tech columnist at the New York Times.

I'm Casey Newton from Platformer, and this is Hardfork.

This week, I go down to Meta Connect to tell Kevin about all the news.

Then, writer and Googler Steven Johnson stops by to talk about the company's hit Notebook LM and how it's using AI to turn PDFs into podcasts.

Finally, has the hat.

It's time for Hat GPT.

All right, Casey, yes.

You just got back to the studio from a trip to Menlo Park

to see some stuff that Meta was releasing.

I'm coming in hot, Kevin.

Yeah.

I hot-footed it right out of that keynote to get here to the studio to tell you about all the latest advancements in the dynamic world of artificial intelligence and augmented reality glasses.

Okay, so set the scene a little bit.

What is it like down there?

What's the vibe?

Tell me what you saw.

The scene at Meta's campus was truly chaos, and I think not in the way they anticipated.

You know, we've been to many of these events.

Typically, when you arrive, you're really only looking for two things, which is like,

where does the rideshare drop you off?

And how do you get into the actual venue?

And usually the way they do that is they have a person standing there and then they also have signs.

Meta, for whatever reason, decided not to do this.

And so people are just sort of wandering around like the duck in Are You My Mother?

And it took me quite a while to get into the venue.

And when I finally found what I was looking for, after, by the way, talking to several Meta employees who, when I asked where Connect was, seemed to have no idea that it was happening that day.

I got a text from a friend who said that that there had just been an accident in the parking lot because their shuttle had hit a car.

So let's just say it got off to a little bit of rough start.

Yeah.

Yeah.

And my invite to Metaconnect must have gotten lost in the mail.

Lost in the metaverse.

Yeah.

Yeah.

They actually sent it as a Facebook message, but no one has checked those since 2018.

No, so I, you know, I am not their favorite journalist.

I've been critical of the company.

They don't invite me to stuff anymore.

But you did get an invite and you hustled your way down there to see what they had had to show off.

I did.

And tell me what you saw.

Well, first of all, should we talk about Zuckerberg?

Yes.

Yeah.

Well, so this is the second time in the past month that I've seen Zuckerberg speak.

And both times he has worn a shirt emblazoned with a classical language slogan.

So he was at the Acquired podcast a few weeks back and wore this shirt that said in Greek, learning through suffering, which he had said was sort of a family motto.

And as we know, this is a person who is obsessed with the Roman Empire.

That is just kind of something that we've known about Zuckerberg for a while.

So, yes, he's very into conquest.

Loves conquest.

His favorite video game is civilization.

Like, this man loves to sort of scheme about winning.

So, Zuckerberg walks out on stage on Wednesday and he's wearing a shirt that says, Out Zuck, out Nahill.

And what this shirt apparently means, Kevin, because I don't speak Latin.

Do you speak Latin?

I speak pig Latin.

Same here.

But it's not a pig Latin shirt.

Didn't say Uckerberg Zay.

I didn't know he said that.

okay

this is going off we're off

today they are coming in hot okay

this can apparently be translated a few ways um it is based on a motto which was out caesar out nihil

which uh was typically translated as like either caesar or nothing or all or nothing but the basic idea seems to be either one that he's sort of putting it on the line sort of win at all costs, or sort of less generously, it's like me or nothing, which is, let's just say, a crazy thing to put on a t-shirt.

Yeah, it's, it's, uh, well, it makes me think that one of his other motives here is since he's gotten rid of his Julius Caesar haircut, he needs to sort of keep some connection to Caesar and his legacy.

If I were him, I would just make a nice Caesar salad.

But I guess that wasn't part of his journey.

So he comes out in the shirt, and I thought that the slogan actually wound up being really apt in this regard.

He spoke for around 40 minutes.

And during that time, he brought out two like creator types, basically, like one was an MMA fighter and one was a sort of more standard creator who like makes cool, fun videos.

But that was it.

He did not bring out anyone else from his team.

You know, he did not toss to any of the other people who work at the company on any of this stuff.

He himself walked through all of the announcements in a way that really, I feel like, was trying to sell the message.

I run this company and I am the Willy Wonka of Meta.

And don't you forget it.

Yeah, it's suck or nothing.

Yeah.

Which I felt was like really aggressive and weird and defensive.

Right.

It's like, it would be weird if Tim Cook did the entire Apple presentation.

Yeah.

Right.

Yeah.

Yeah.

Okay.

Aside from the sort of, you know, semiotics of Mark Zuckerberg appearing in this

Caesar homage t-shirt and not letting anyone else onto the stage with him.

What else did he actually announce?

What were the products and the sort of updates that they released?

Sure.

So I'll sort of do these in an ascending order of importance if that works.

Okay.

So first of all, they have.

It's like they always say, start with the most boring stuff when you're doing a podcast.

Well, I feel like if we do the reverse, then the episode gets more boring the longer I go on.

Do we want that?

I trust your vision.

Okay, so I trust your vision, which is something that you should do if you're wearing a VR headset because you're being lied to.

So the first thing they introduced is the Meta Quest 3S.

This is an entry-level headset and it sells for $300.

And until this week, Meta's original Quest 3 was selling for $650.

So this is sort of a big step down.

They want to get more people into VR.

We'll see if that works.

They also put out a software update for their Ray-Ban Meta smart glasses.

And they're able to do things like you can set a reminder or you can scan a QR code.

And later this year, they say you're going to be able to do real-time language translation.

So you could be speaking to me in pig Latin and then I would be able to understand.

understand it because I would sort of be getting a real-time translation through the speakers in the arms of the glasses.

At the anse

interesting, yay.

Yeah, so if you were able to understand that, you're probably living in the future because you have the new glasses update.

So there were those things.

And then there was just like a bunch of AI stuff, including most interestingly, and I have to get your take on this.

They're going to start putting fake AI slop into the feeds.

Did you see this part?

No.

Okay.

So now they say it's just a test, but as you browse through the Facebook feed and the Instagram feed in the coming months, you're going to see stuff that is based on your interests or has your likeness.

So if you live in Chicago and you've posted content about basketball, you might see a picture of yourself in a Chicago Bulls jersey.

And then you can swipe on it and you can see sort of additional like you plus Chicago Bulls related content.

And I'm dying to know what you think about that.

So well, sorry, are these ads for like Chicago Bulls merchandise?

These are, these are just organic posts appearing in the feed starring you, the user, in something related to something that you've posted about or clicked on recently.

Exactly.

Exactly.

So they're looking about what you're posting about or the sort of things that you typically view.

And they're saying, can we make synthetic stuff out of that and show it to you?

And will that be more engaging to you than hearing from your actual family and friends?

Oh my God.

I mean, this is...

the creepiest thing that I can imagine them doing.

Like people already believe erroneously that Facebook is listening to them through their phone and like tailoring targeted ads to them based on like conversations they've had in their house.

With this, I mean, imagine you are, you're talking about fishing with your friend.

Um, and, you know, all of a sudden, because you've clicked on some fishing stuff, you're just scrolling through your Instagram feed and you see a picture of yourself in a fishing outfit going fishing.

Like you are going to throw your phone into the nearest body of water and you're never going to log on again.

I mean, again, and so I talked this afternoon with Chris Cox, who's a sort of high-ranking product executive at the company.

And he said, basically, like, Casey, like, do not overstate this.

We're going to be very gentle about the way that we are rolling this out.

And yet, at the same time, Kevin, a year ago, I wrote a post called The Synthetic Social Network is coming.

And it was based on this idea that eventually they were going to realize they could probably show you something more engaging that they just made up than something from your actual family and friends.

And while I'm sure there will always be a mix of those two things in anything that they make, we are arriving in this world where the app on your phone is just going to be showing you made up AI stuff.

Yeah, I think this is a really good point because it marks a really big departure for Meta.

You know, this company has historically been about connecting you to people that you know in the real world, whether it's your college classmates or, you know, people you work with.

Like real friends and real family has always been the kind of lifeblood of these services.

And I think what we've seen over the past few years is that they are just not aiming for that at all.

I mean, now, you know, I'm assuming some huge part of the usage of Instagram is just reels, right?

It's just, you know, content from people you don't know, from influencers, from people outside your network who are posting stuff.

It seems like they have basically given up on the task of building a social network around your real friends and family.

Maybe your real friends and family don't post that much anymore.

And so instead, all they have left to show you is kind of disconnected social content from influencers and people who you don't know, but might make interesting things, and this kind of AI generated slop.

Yeah.

And, you know, I'm sure people are going to have different experiences based on when they started using these products.

But for you and I, who started using Facebook in particular near the beginning, this world where we increasingly see nothing that friends and family are sharing because they're just mostly not doing that anymore.

And we're seeing sort of AI generated images and videos.

I just think Facebook is going to start to feel like an abandoned amusement park, you know, where it's like a little bit creepy.

You know, you feel like they're about to put a spirit Halloween banner at the front of it.

Like it, there's just something that feels so strange about this moment.

And while I take Chris Cox at his word that they are going to be careful about the way that they do us, they're telling us where they're going.

And I have no reason to believe that they're not going to push very hard in that direction.

Because for the longest time, the biggest problem that social networks have is that they just kind of die by default.

They're novel and fun for a while, and then people just kind of get tired of them and they move on.

And, you know, Facebook did an incredible job of keeping people entertained there for a very long time.

But if that is starting to wane, particularly in the United States, what better solution than to say, you know what?

It doesn't matter that your friends and family aren't posting anymore because we can just make up infinite content to show you.

And Facebook and Instagram become apps that are just about whatever will keep you looking at them.

All right.

What else did they show off?

Yeah.

So this gets to the last big thing that they showed off.

And this really was the star of the show.

And it's called Orion.

Now, last week, Snap showed off their version of this, their latest spectacles.

These are augmented reality glasses.

And the basic idea is that inside a pair of glasses that you wear on your heads, there are these miniature projectors called waveguides that will project images in front of you that look like a computer operating system.

And you can then interact with that operating system using your voice or using hand gestures.

So you might use it to watch a video or play a game or make a video call.

So I saw the snap version of it last week.

This week, Meta came forward and said, Here's what we've been working on.

And as far as I can tell, Meta is much further along than Snap is.

Yeah, when I saw Snap's version, I almost couldn't believe how ugly it was.

It looked like the sunglasses that they give you after you go to the eye doctor, where it's like the blocky plastic sort of goggles.

That's what Snap's goggles look like from the outside.

I haven't tried them on yet, so I don't know what they look like when you put them on.

But did Meta's version of this, their Orion glasses,

actually look better?

They are chunky and funky like the snap glasses are.

I'll sort of leave it to the listener to decide which pair they think looks better.

But it seems like the technical capability in these glasses is stronger, at least in talking to some of the people who have used them and reading some of the coverage today.

I'm hoping to get my own demo soon.

But for example, the field of view, which is essentially the area in your vision that can display the operating system or any sort of digital elements that you might be interacting with.

The snap glasses have a field of view that is 46 degrees.

And this to me was the worst thing about these glasses, Kevin, because every time I move my head, the interface would disappear.

It was like I was playing peekaboo with an operating system, right?

The meta field of view is 70 degrees.

And so most of the objects that you would be interacting with just sort of stay in the frame for a long time.

Now, like what I'm envisioning something like the Apple Vision Pro, which does allow you to do things like pull up your desktop in your sort of headset or play a video or something like that.

So how are these different from something like the Apple Vision Pro?

So those technologies are similar in some ways, right?

They are trying to get to the same destination, but there is a belief within meta that Glasses are a much better and more natural ultimate form factor for the next generation of computing than a VR VR headset would be.

And the reason is that a VR headset is giant, it's bulky, it's distracting, right?

You're sort of always aware that you're wearing it.

And in my own personal experience, whenever I put one of those things on my head, within a half an hour or so, I want to tear it off and not put it on again for a really long time.

Glasses are different.

You're wearing glasses right now.

My guess is you have not even thought about the fact that you're wearing glasses for the past half hour.

Yeah, well, I am watching a YouTube video while I'm talking with you in my glasses, but these are very advanced headsets.

How dare you?

How dare you?

So Meta believes that the final destination here is glasses that are as good at everything that the Vision Pro does, but in a sort of pair of lenses that sit easily on your face, are not bulky, do not make you nauseous.

So that's where they're trying to get.

Now, can people go out and buy these

glasses that Meta is showing off, these Orion augmented reality goggles?

No.

So Meta is made about a thousand units and says that they're going to be mostly using them internally, but they are going to make some some units available to external partners, presumably app developers, and other people who might want to try them out and see what they could do with them.

But you know, Kevin, if you want to know why aren't they selling this directly to consumers, Alex Heath has had some great reporting in The Verge this week about how Meta did initially want to put these things on sale.

But among other considerations, they could not get the cost per unit below about $10,000.

Oh my God.

And Kevin, I know what you're saying.

You're saying, Casey, why are they so expensive?

Well, these glasses are not truly standalone glasses, which I should say the snap glasses are.

This prototype has what they call a neural wristband, which is a little loop that you put over your wrist that helps the glasses to interpret your hand gestures so that you can control the operating system.

And there's an external puck that processes a lot of the computing.

It's like a hockey puck?

It's sort of more

tube-like.

And

the tube puck.

You know know what?

You're actually making a great point.

And if I were them, I would not have called it a puck.

If you try to play hockey with this thing, you're going to be sorry.

Also, it costs $10,000.

So please don't hit it.

Such a good point.

So.

Okay, so they're sort of offloading some of the, because one thing that Apple has talked about is just how hard it is to shrink down something like the Apple Vision Pro smaller than it is, because it requires a bunch of processors.

It requires a big battery.

So Apple actually made the battery pack sort of external.

So you have to carry that around.

And it requires cooling and all this other stuff.

So

is this puck part of how Meta is trying to shrink down the actual glasses by just giving you more things that you have to carry along with the glasses?

Yes.

So you're exactly right.

And it's, again, it's fascinating how a device that is being positioned as something that will be more convenient for you and will be this incredible assistant and so much more effortless than wearing a giant headset comes with a wristband that presumably you've got to charge and a puck that you've got to charge, right?

It's, you know, you have two peripherals for the pair of glasses on your face.

So, you know, I've just been reflecting recently, Kevin, about how

this story that we're discussing right now, the hunt to build great AR glasses has been going on for more than a decade now in Silicon Valley.

And this week we saw the state of the art.

And the state of the art is it costs $10,000.

You have to have a peripheral to control it and you have to do the computing on a separate device.

Okay.

And that is after tens of billions of dollars that have been invested.

And you just compare that to something like artificial intelligence, which we talk about so much on the show.

And you look at the leaps that they have made in the past two or three years.

And I believe it has left us in a situation where what we've just been discussing, the huntable these AR glasses, is the hardest problem in consumer technology.

I mean, to be fair, the building and training of large AI language models has also required tens of billions of dollars of investment in all of these GPUs from NVIDIA.

So in some sense, like there was no sort of cheap and easy way to solve the problem of putting a high-performance computer on your face.

Right.

And that gets into, well, you know, why is it that these things have developed at such different rates?

And why is it, Casey, that these things have developed at such different rates?

I'm so glad you asked.

Because AI is a pure software problem.

The hardware problem of AR is just much trickier because you're running up against the laws of physics and about unsolved problems in battery life and how do you dissipate heat and how do you shrink down microprocessors and optical stacks and everything else.

And so it is just a super hard problem.

And while every year it seems like we get two or three percent of the way toward the goal, at the same time, again, the state of the art is something that you will not be able to buy in stores and like frankly would just not be all that good if you could.

Yeah.

So I understood understood when Apple launched the Apple Vision Pro why they were making

products in this area, right?

So they knew they could trick you into spending almost $4,000 on one.

Yes.

Well, no, they're a hardware company, right?

That is where they make their money.

They

know that smartphones are not going to be around forever.

The iPhone is not going to be the final evolution of personal computing.

And they see these head-mounted displays as kind of the successor to maybe something like the laptop.

What's less clear to me is why Meta is investing tens of billions of dollars into developing this stuff, right?

There's obviously one story that we've talked about before, which is they're very sick of building things on Apple's platforms.

They don't like having to work within the constraints that iOS developers have to work on.

So they want to own the next hardware platform, but I guess I don't see it as being a natural evolution of their social media app product strategy.

So can you give any sense of why they feel like this is so important for them that they're going to spend billions of dollars developing these glasses?

Well,

the ultimate reason is that they believe that if they don't, someone else will.

And this has always been Zuckerberg's approach to building products, right?

He is deeply paranoid, you know, and not in a sort of conspiratorial way, but just in a sort of very observant sense that when you look at the technology companies of the past 50 years, you know, the companies that were biggest 40 years ago, many of them are no longer around or are nowhere near as dominant as they once were, right?

So, Zuckerberg knows there's always a wave that is about to hit you and potentially from behind.

And the best way to avoid that happening is that you go out and you invent the future yourself.

You know, you mentioned their feud with Apple, and this is it's deeply personal to him.

He mentions it more and more in his public appearances.

And, you know, you saw the shirt.

It's Zuck or nothing.

And a Zuck or nothing world does not have room for Apple in it.

And it is a world where Mark Zuckerberg himself controls all of the most important computing platforms.

So, you know, to find out how things are going, we'll just have to see what slogan appears on his next t-shirt.

When we come back, a version of the notebook with a much happier ending.

And now, a next level moment from ATT business.

Say you've sent out a gigantic shipment of pillows and they need to be there in time for International Sleep Day.

You've got ATT 5G, so you're fully confident, but the vendor isn't responding.

And International Sleep Day is tomorrow.

Luckily, ATT 5G lets you deal with any issues with ease, so the pillows will get delivered and everyone can sleep soundly, especially you.

AT ⁇ T 5G requires a compatible plan and device, coverage not available everywhere.

Learn more at ATT.com/slash 5G network.

At Capella University, learning online doesn't mean learning alone.

You'll get support from people who care about your success, like your enrollment specialist who gets to know you and the goals you'd like to achieve.

You'll also get a designated academic coach who's with you throughout your entire program.

Plus, career coaches are available to help you navigate your professional goals.

A different future is closer than you think with Capella University.

Learn more at capella.edu.

Well, Casey, we talk a lot about AI tools and products on this show, and I have one that I'm really excited to talk about today.

Yeah, I'm excited to talk about this too.

This is a new product called Notebook LM.

It is a tool from Google, and you can think of it like a kind of personal research assistant.

It's a piece of software that allows you to upload documents, PDFs, word files, even audio files, websites, whatever you want, into these things called notebooks and then use Gemini, Google's AI model, to basically chat with the documents, to sort of have a conversation, to ask questions, to create study guides or summaries.

You can even use it to create a podcast about the material that you've uploaded.

Yeah, and you know, this was a product that was announced at Google.io.

They called it Project Tailwind back then.

And from the moment I saw them talking about it on stage, I thought, I have got to get my hands on this thing.

And while it is true that you can use other tools to chat with documents, that is not unique to Notebook LM.

They have really focused on making sure that you can cram as much material as possible into their system so that you can have conversations with not only one very long document, but many very long documents.

And that has really been the difference for them.

But as you note, they recently came out with something that is maybe even more impressive.

Yeah, the AI audio overviews feature is really what has been sort of lighting up the internet over the past week or so.

Like, I'll just be honest, we see a lot of AI products.

We get a lot of early access and demos of things.

And many times they show some potential.

But if you actually start digging in, they are not all that useful or they hallucinate or they're just not reliable enough to be useful for people like you and me.

Notebook LM is, I would say, one of my favorite AI products that I have used this year because it is not trying to do everything for everyone.

It is a tool for research, for writing.

It is really, really capable at what it does.

And the audio feature in particular is just pretty stunning.

So today we're going to bring in one of the key people who helped conceive of and build Notebook LM, Stephen Johnson.

Steven's path to working at Google is pretty unusual.

His main career, the thing that he's most known for, is a writer.

He's written for many years for the New York Times and New York Times Magazine and other places.

He's the author of more than a dozen books, including his latest, The Infernal Machine, a true story of dynamite, terror, and the rise of the modern detective.

And he's been one of my favorite writers about tech and the future.

And a few years ago, Google approached him and basically said, hey, want to help us make a tool for writers?

Yeah.

And, you know, I got to know Steven a little bit as they were launching Notebook LM and we met and he told me all about his note-taking process and how he wanted to use AI to sort of improve his writing.

And we truly just became fast friends because we have the exact same view of this stuff, which is give me the most technology to make my job as easy as it possibly can be.

And unlike me, he's now actually working inside this company trying to make something that does just that.

Yeah, and why haven't you written 14 books?

Remind me?

Well, you know, I've been busy lately, Kevin, but I'm going to get around to it one of these days.

Let's bring in Steven.

Stephen Johnson, welcome to Hard Fork.

Guys, it's great to be here.

Hey, Stephen.

So, Stephen, I remember reading a piece you wrote for the New York Times Magazine back in April of 2022.

It was about six months before ChatGPT came out.

And you had this big, great piece about how AI was starting to get really good at language through these new things called large language models and your sort of predictions about how that would have all kinds of profound effects on society.

But I remember that piece so vividly because it captured this feeling that I was having at the time, which Casey now has started calling AI vertigo, basically this sort of head spinning sensation when things are just moving so quickly.

But I'm curious, like, what got you as a writer interested in AI to the point that you decided to sort of build AI products at Google?

You know, I just have spent all of my career as a writer always dabbling with tools to help me do the writing with with all the latest software like casey and i have this kind of shared obsession with note-taking software oh you're one of those you know i know i know you know i don't even get me started about scrivener and all the different things that we can talk about but i always saw the computer and software as a kind of companion and a kind of brainstorming partner like and i was always pushing the technology to do that in my own work and so

the idea that i could just kind of say hey let's think out loud about this particular topic and it would understand on some level and respond with coherent sentences.

You know, obviously there are hallucinations and there are all the things that we know are problems, which I also wrote about in that piece.

But it was clear that some new

set of doors of possibility had just opened up for the first time.

And I just got really interested in walking through those doors.

So how does Google first approach you?

And is the idea, hey, we want to make a tool for writers like you or was it something else?

Yeah, it was a little bit like that.

So Google had just spun up right around this time, a new division called Google Labs.

There was an old Google Labs.

This is kind of a new iteration of it.

And there was a guy running it named Clay Bevor who's since left.

And now Josh Woodward is running it.

And Clay and Josh had this idea that Google Labs could be a space where you could do kind of product-focused experimentation with new emerging technologies.

And they also had this idea that there would be...

co-creation would be built into the kind of ethos of labs.

And so if you were making a music product, you would have a musician in the room, like as you were building it.

So it wasn't just about, you know, we're going to go out and do some user research with musicians.

We're actually going to have somebody through the life of the product there.

And so they were just kind of cooking up these ideas.

And they had both read my books over the years.

They'd read that timespiece.

They'd read my sub stack.

And all that together caused them to think, wonder if Stephen could be the first kind of guinea pig for this.

So, you know, Stephen, you and I have had a chance to chat before.

And

I truly aspire to be the note-taker that you are, because I have talked talked a big game on this podcast about how I'm trying to write down sort of every interesting like a quote or idea that I come across and link those together.

And I've made some strides there, but like you showed me your system at one point and it is the real deal.

Like you truly have been keeping track of every idea that you've come across for seemingly quite some time.

What was the moment that you said, oh, like this intersects with AI in a way that maybe this notebook LM can realize in a product?

I want you to remember, Kevin, that every time you think that Casey is such a super nerd with his note-taking, like there are even more deranged people in the world.

You are the alpha nerd of the note-taking community.

So

yeah, I can tell you exactly what it was.

So I have been collecting quotations from books that I've read.

Initially by typing them up in the late 90s.

And then once e-books came out, you could save quotes and things like that.

There's an amazing program that I think you use called ReadWise.

It lets you organize all your quotes from, if you read on the Kindle or any other e-book.

And so I have something like 8,000 quotes from books that date back to the late 90s that I've collected.

And that is really the history of all the ideas that really shaped who I am, right?

Like my mental model of the world is shaped by the other ideas that I've read from other people.

And so Notebook now lets you have

up to 25 million words in a single notebook.

Put that in terms of pages.

How many pages is that?

What would that be?

That would be like 40 books.

Yeah.

Right?

Something in that order.

And do your 8,000 quotes fit in one of these 25 million word documents?

Yeah, yeah, there's normally about three main words.

So I've loaded them all in as a bunch of documents.

So 30 years of collecting quotes fits easily into one of these notes.

Yes.

And what I'm slowly adding to that notebook is all the stuff that I've written, too.

So it's kind of everything I've read that's important and literally every word I've published is eventually going to be in that one notebook.

I just haven't seen it.

Kevin can't do that because it would poison the data set.

I think it's good.

It works for for you.

The safety flag.

So it would be going off.

It'd be terrible.

So

when I was able to do that, which was really, I don't know, about a year ago for the first time, or I can get all that stuff in there, I call that notebook my everything notebook.

And then I could sit down and just be like, I'm thinking about writing a piece about X, or here's a paragraph in the piece that I've just written.

What am I missing?

What am I forgetting?

Give me an overview of all the stuff that I've read that is related to this particular topic.

And it would return, particularly once we switched to Gemini.

Like Gemini Gemini was the big kind of paradigm shift here.

I get this like incredibly nuanced response that is constantly reminding me of things that I've forgotten.

And now, as of like three months ago, we have inline citations in all the comments from the model.

And you can click on each citation and it takes you directly to the original quote.

Yes, I love this feature.

I've been playing around with Notebook LM, and it is truly one of the best features about it is that it'll show you something.

You're talking to it about about something you've read or something you've written.

And it just has that little sort of like citation.

You click on it and it takes you right to the source material.

So you can see for yourself, like, this actually is an accurate representation of what was in the PDF I uploaded.

And

what an incredibly interesting learning mode that is, right?

Like up until now, if you wanted to have a conversation with...

the material in a book, like you had to find the author or you had to find a tutor and an expert who knew the material really well.

And those people are in scarce supply.

But now you can actually like load in the book and navigate it through conversation and dialogue, which is a form that people really obviously like to use.

So that is amazing.

And I think will probably be the primary way that people access it.

But because of the fact that you have taken these notes for 30 years, you're able to use this in this different way, which is essentially like, take me through my own intellectual history and let me talk to the entire like course of learning that I've had over three decades, remind me of things, right?

Or make new connections for me.

And that feels like the kind of augmentation that I truly have always wanted AI to give us, right?

Like that is the good stuff.

Right.

I'm glad you said that.

That means a lot coming from you.

I think there was an early tension in

creating Nopokellum, which was the question of like,

How normal am I?

You know, we have this amazing colleague, Rosa Martin, who's the product manager, who's been incredible, was on it from the beginning.

And I think like, you know, she, she was kind of like, this has to work for people who don't collect 8,000 quotations over 30 years of their lives, right?

If it doesn't, you know, work for them.

And so, but I think one of the things that we've learned is you actually, you know, particularly in a digital age, like you have, you know, you can import docs and slides.

And so if you're a Google Drive user, the history of all the docs you have in there is actually a history of the last four or five years of things that you've been interested in and that you've been working on, whatever your job is.

And so one of the things I often tell people with Notebook LM is just first thing to do is create a notebook.

If you are a drive user, grab the last 20 docs in there and just don't even think about organizing them.

Just dump them all in, even if they're for different projects, and then just start having a conversation.

And the sense of, oh, this AI actually knows what I'm doing and understands what I've been working on and can piece together kind of, you know, insights from is pretty amazing.

And by the way, I should say, you know, when we do this, we're not training the model.

That was one of my questions.

I think a lot of people would say, okay, well, if I upload a copy of my book or some documents that are personal to me, is Google going to then be able to sort of see and train on that material?

So what we are doing is not training the model.

That would take a long time to train a model on your data.

And it actually wouldn't, for complicated reasons, work as well as the way that we're doing it.

We're just taking the information you have and putting it in the model's context window, which is kind of like the short-term memory of the model.

It's the easiest way to understand it.

And the beauty of that is, one, the model is much more accurate with information in its context.

And so the hallucination rates drop down dramatically.

You can do things like citations that you wouldn't be able to do otherwise.

But it also means that the second you close your session, that information goes away.

And so there's no way for we're not learning from it.

We're not making the model smarter in the long run.

And there's no way for that information to kind of leak out into other users.

And that has been a fundamental principle of the product from the very beginning.

Yeah.

I hope you take this next thing that I'm about to say the right way, because I do mean it as a compliment.

But notebook LM strikes me as an extremely un-Googly-y product, right?

It is probably not the kind of thing that's going to get a billion users, which is how Google has historically decided what to build.

It didn't have like a big splashy launch with ads running on the Olympics.

It's not sort of promoted on a bunch of other Google products, as far as I can tell.

It has a Discord server.

And the design of the actual tool just feels different than a lot of what Google has built in AI.

It sort of feels to me like it might be this kind of isolated, kind of semi-autonomous region within Google that doesn't have that much contact with the rest of the company.

Is that right?

Well, that's an interesting question.

Some of that is right.

And some of that is a reflection of what labs set out to do, right?

Which is to like, let's create a space where we can be more comfortable with being experimental.

And that enabled us to do some things like

experiment with different types of interfaces that would not necessarily have the polish that you would expect from other Google official products.

The Discord is a great example.

That was one of Rise Martin's ideas.

You know, we just wanted to like build a community around it.

And I remember Riza coming to me and saying, you know, I want to build a Discord for this product.

And I said, what is a Discord?

I had no, my kids were Discord users.

So it rang a bell, but I'd never been on Discord before.

And now I'm in there all the time.

And we have, we have like 45,000 people who are members of this community now.

And, and we just discover so many things from them.

Like it's like notebook is taking off with DD players, like Dungeons and Dragons, like Dungeon Masters, because they have, it's a very like literary genre of game, right?

And you have these long campaigns.

With like rich lore.

Yeah.

Yeah.

Yeah.

And same thing with like fantasy novelists and sci-fi novelists where they have a backstory that's enormous and they can't keep track of it all.

Like if George R.

R.

Martin would adopt our product,

we would have

to do it.

I spent some time in the in the notebook LM Discord.

It's a very fun place because you get to sort of see how people are using it and sort of get ideas.

I also love that there's one person in there who's just constantly posting about how they're using Notebook LM to analyze a huge database of Sonic the Hedgehog fan fiction.

You know what I'm talking about?

They've been there for a while.

It's really interesting.

They found a lot of use cases for Sonic-related

work.

And they're going to be on the show next week.

But you know what, Kevin, I want to disagree with you about whether this is Googly or not, because I think this is like old school Google.

And this is the Google I like.

Like, do you remember back in the day when the Googlers could just do anything they wanted in their 20% time?

They give them a day a week to be like, hack around on something, right?

Do something interesting.

To me, this is the sort of thing that would come out of 20% time where it's like, let's find some of the, you know, the biggest, everyone at Google is a nerd, but let's, let's find people who are nerdy about something really in particular that could be massively useful to maybe a narrow group of people.

But maybe we find something in there that who knows does scale up to a billion people eventually.

Yeah.

But maybe we should talk about the audio, Kevin.

Yes, we have to talk about the audio feature because this is what really has put this

tool on the map for a lot of people.

This is an amazing feature.

The first time I saw it, I did have a moment of AI vertigo and I think I emailed you and was like, oh my God, what is this thing?

How is this feature so good?

How does it work?

What is it trained on?

How did it learn how to do podcast banter?

Just tell us about this feature.

It's another great labs case study.

It was another team inside of labs that had basically developed a tool that would take any source material you wanted and generate an audio conversation that would sounded like two engaged, entertaining people having a conversation, you may be familiar with this genre,

about whatever material you gave it.

And the kind of two use cases that we were talking about in the early days was

kind of source material that

no one would ever build a real podcast for.

So arcane city council meetings that no one, there's no economics in turning that into a podcast, or personalized learning where you're like, you're an auditorio learner and you want to, you know, do a review of the week's, you know,

assignment and you'd rather digest it in the form or you'd like to augment it with the listening to a conversation because people remember better with conversations and they can do it on the go.

And so they had this incredible demo.

And the thing about it, like behind the scenes, is that a lot of the breakthrough is actually the edit cycle.

So behind the scenes, it's basically running through

stuff that we all do professionally all the time, which is it generates an outline.

It kind of revises that outline.

It generates a detailed version of the script.

And then it has a kind of critique phase and

then it modifies it based on the critique.

And so

it takes about four or five minutes to generate.

And it's because it's going through all these different passes.

And

you can call that like chain of thought reasoning.

But when I saw it, I was like, no, no, no, that's an edit cycle.

Like that's where you did a draft and then you revised it and then you got it got better over time are you taking notes you could do this for your book

it would help go ahead Stephen so and then at the end of it there's a stage where it adds my favorite new word which is disfluencies so it takes a kind of sterile script and turns adds all the banter and the pauses and the likes

the following words

and

that turns out to be crucial because you cannot listen to two robots talking to each other.

No one, it would be just painful to listen to.

It'd be like, I don't know, the Lex Freeman podcast is pretty popular, Steven.

Yay!

Oh, boy.

So,

I'm not even going to bring that up.

Follow up on that.

So, that was crucial.

And then, on top of that, all there's some new voice technology, which adds

really an incredible layer, which is like figuring out without any coding in the script, figuring out that this is a point that they are trying to emphasize.

And so they are speaking more slowly or they're trying to imply that they're hesitating a little bit.

And so they're raising their intonation a little bit.

All that stuff, it does.

And having two people talking to each other like that, it just, it is one of those moments.

When I first heard it, I was like, this is incredible.

And we were already, we were already, we rolled out these notebook guides that take your, all your documents and turn them into a briefing doc or an FAQ or a timeline, which is incredibly useful for

writers.

And so this was just like, oh, we can now do it in another form.

Like if maybe you want to take your sources and listen to a conversation about them.

And so it just was a beautiful fit inside of Notebook LM.

And so we just have been scrambling all summer to get this out.

And it's been really cool to see this.

It's really awesome.

I made a podcast about my new vacuum cleaner that I got by feeding it the user's manual PDF and how it pops this eight-minute explanation of all the features of my new vacuum.

It's really cool.

Okay, so I have a confession, which is I am obsessed with this stuff.

And Stephen was very kind and gave me early access to this.

But at the time, I was, you know, getting ready for Meta Connect and some other things.

And I just did not have the time and attention to focus on it.

But then when I found out when you were coming on the show, I thought, I am not going to listen to any of these until we're all in the room together.

Oh my gosh.

And here's, and here's why, because I've learned from YouTube that the most popular thing that you can do on YouTube is to hear something for the first time.

So, you know, I don't know if you've seen this, Kevin, but it's like if you listen listen to Metallica for the first time on YouTube, you got a million views.

Yeah.

But I already listened to Metallica in high school.

So I thought I'll do the next best thing and listen to the notebook LM audio right here on Hard Fork.

So, all right, so let's listen to a few examples of this audio feature.

I have been playing around with this for a couple of days now, having a lot of fun with it.

So, a lot of times in our work as journalists, we have to sort of make sense of a bunch of different documents, whether they're legal filings or what have you.

And so, I was doing some research about Waymo and their self-driving cars.

And there have been a few studies that have come out recently about the safety data of these cars, of human drivers, sort of how safe are Waymos compared to human drivers.

And it's been a really hot topic we've talked about on the show.

Totally.

Very controversial.

But it's a little hard to understand.

The data is a little mangled, and they're just, these papers are quite long.

And so this morning, I was going into the studio and I thought, I'm just going to dump a whole bunch of these PDFs of these studies into a notebook LM and generate a a podcast that I can listen to on the way to the office and maybe get a sort of high-level overview of what these studies have shown.

So I want to play for you the first sort of 30 or so seconds of my Waymo data podcast.

All right.

Ever see one of those Waymo cars just cruising around with like no one behind the wheel?

Yeah.

I always wonder, is that thing safe?

I mean, no offense to robots or anything, but handing over the keys to a machine, it just feels different.

Yeah, it really does make you think about like trusting technology with our lives.

Exactly.

Especially when it's something as important as like driving, you know?

Totally.

So that's what we're diving into today, the safety of those driverless Waymos.

We've got a bunch of research lined up,

including some really recent data to get past the headlines and figure out what's really going on.

So that's amazing.

That's one clip.

That's amazing.

It's really cool.

Okay.

And it does actually continue on for like minutes after that and sort of break down the data in these these papers quite well from what I can tell.

I also put in your latest platformer newsletter into this and had it generate a podcast about that.

So here's a clip from the platformer newsletter AI podcast.

Well.

Speaking of risks, Newton's decision to leave Substack, that was a risk.

But it sounds like he's thinking long term.

He is.

He really is.

And I think that's key.

Building a sustainable media business in this day and age.

It's not easy.

It takes more than just great content.

You need smart business decisions, too.

It's a balancing act, right?

You got to stay true to your vision, but also make sure you can keep the lights on.

And Newton's been very open about platformers' finances, about the challenges that come with going independent.

Transparency builds trust.

And these days, trust is invaluable.

And let's be real, leaving a platform like Substack, even one with its problems, it's going to come with some financial growing pains.

For sure.

It's like jumping off a cliff and hoping you can build your wings on the way down.

I like these people.

They're smart people.

They know what they're talking about.

Yeah, they're big fans of you.

The last clip I want to play, I was just sort of thinking, like, how esoteric can I get here?

Like, what can I make a podcast about using Notebook LM?

So I uploaded my most recent credit card statement to Notebook LM and had it try to make an AI podcast about some of the things that I've been spending my money on.

And so this is the

most recent credit card statement that I got in podcast form.

I love it.

Okay, let's see.

I'm noticing a pattern here, quite a few Uber rides between August 8th and September 9th.

Yeah, and that's something to consider, right?

Especially if you live in an area with readily available public transportation or bike-friendly routes.

Those real quick ride shares can really add up.

For sure, they do.

I mean, for example, let's say an average Uber ride costs you $20 and you're taking four of those a week.

Well, that's $320 a month.

Think about it.

That's money that could be going to other things, other financial goals.

Yeah, absolutely.

Small changes can make a big difference.

Yeah.

It really told me to get my ass on the bus.

It's like a financial advisor in your pocket.

Truly.

So this

blew my mind.

What's your reaction to this?

It really is extraordinary.

And, you know, again, I knew we were going to do this today.

I wanted to wait until this moment to hear it.

But in the meantime, I was seeing so many folks on social media saying, you have to listen to this.

Like it is so eerily good.

And, you know, my mind is already alive with, you know, a problem that I have, which is because we do a podcast, we often talk to the authors of books.

And often we decide we want to talk to them, you know, a week before.

And then I get a PDF

in my inbox.

And now I have seven days to read it.

And it's incredibly difficult.

If I could listen to a podcast about it, I would love it.

Now, I realize how painful that's going to be to hear for every author

who exists.

But,

you know, if it makes information more accessible, I want to try it.

Yeah.

And probably people will make podcasts about a lot of the books that we talk about in this show, but no one's going to make a podcast about my credit card statement or

the policies and my kids' daycare or something like that.

So it's a really interesting way to sort of transform these sort of more esoteric documents.

I mean, once I heard about your profligate Uber spending, I do kind of want to do a podcast about your spending habits, but go on.

Well, I think actually that one was really interesting because one, its ability to figure out a way to rationalize a podcast is amazing.

We would give it kind of internal notebook LM documents, and they would be like, well, it's really exciting.

We've gotten our hands on some internal documents from Google.

And, you know, they just, they want to turn it into something that's a show.

But

what you could hear there is an interesting thing.

They are generally

instructed to be enthusiastic and engaged.

So one thing people are doing is putting their CVs and resumes in there and they're like, John Smith, I mean, what an amazing career he's had.

I mean, assistant vice president at the bank.

That's impressive, you know.

But what happened there, apparently, because your Uber writing is so excessive, was an interesting

suggested mode, which is critique.

Yes.

Right.

So you can imagine a future version, which is like, I actually, I want some tough love here.

Like here, here's the thing I'm working on.

Like, talk me through like what the problems are.

I want to hear that.

And that could be something that we can do.

And it's amazing because I didn't give it any prompt.

I didn't say

criticize my spending on Uber.

It was literally just make a podcast out of my credit card statement.

I uploaded the PDF.

I pushed a button.

I waited a couple minutes.

I listened to this podcast and it had pulled out some details and spending patterns from my credit card statement.

It's just an incredible.

Are we seeing the seeds of our demise as podcasters in this segment?

Well, that was the second thing I was going to say, which is that they're instructed to, and my mind is kind of changing on this as users are experimenting with it more, but they're instructed to be fun and engaging.

And I listened to dozens and dozens and hundreds of these, you know, as it was in development over the summer.

And

they were always fun and engaging, there's banter and all that stuff.

I never once heard them be funny.

I never once laughed.

And I thought about,

you know, since you've been saying so many nice things about me, I thought about hard fork, which I laugh at out loud all the time when I listen to you guys.

And it occurred to me like, oh, the one thing, interestingly, that the models, they're so good at so many things, so almost, you know, superhuman in some of their abilities, but they can't yet be funny in this way.

So, you know, who knows?

Obviously, being able to push it in directions and give it some guidance is something we're getting a lot of requests for.

So I would say that's a good question.

You could also imagine being able to sort of pick, like, do you want this to be a two-person sort of hosted podcast?

You know, or do you want it to be just one person?

Do you want, do you want someone to, one of the hosts to be like super annoying and like always interrupting?

You could call that Casey mode if you wanted to.

That's a free word.

That reminds me.

I think we should do Kevin's credit card statement, but as a true crime serial, you know?

So do you do any kind of moderation on what people are uploading into Notebook LM?

Could someone upload

a Mein Kampf and make an AI podcast about it?

So there's basic safety, low-level safety that Google has that will block really offensive things from happening.

And basically, what we've,

the kind of the latest version of it,

when there is politically charged content from the either left or the right,

the hosts are instructed to take a neutral kind of reporting tone and to basically say, we're not taking sides in this, we're just reporting what is in these sources.

And we feel like that's kind of the best way to do it.

And so there's kind of an extra instruction for them to make that clear

if it seems like it's politically charged in some kind of way.

Yeah, I did actually try making an AI podcast about Mein Kampf.

That wasn't a hypothetical.

And what happened?

It did it, but it was very judgmental at the top.

It was like, we're going to talk today about Mein Kampf, a book about how Hitler became such a monster, which I actually thought was good.

Which is actually how we teach history, by the way.

Yes.

You know, people do read Mein Kampf in history classes, but someone's up there telling them that it was bad.

Totally.

So in Google's press release about this feature, there was a brief mention of some other features that might be on the horizon for Notebook LM, including the ability to generate overviews, podcasts in a language other than English, and the ability to actually take part in the conversation yourself, to sort of interrupt the host of this AI-generated podcast, maybe ask a follow-up question about something they just said.

How far away do you think those things are?

We're actively working on both of them.

I would expect some version of them to be in the next few months.

We actually just

this week allowed you to input audio as a source too.

So now you can go like this kind of classic format in the student use case, you go record your lecture, take handwritten notes just at the most important things, and then you go back to notebook, upload the audio lecture, and then basically say, take my little high-level summary and expand it.

based on the full text of the lecture.

So I can really just write down the most important things and let the model fill in all the details based on the recording and then take that and turn that into a podcast so I can review it at the gym.

Like that sort of workflow, like

I would be so good at college if I went to college today.

I mean, truly, I truly think that I could be at the top of my class.

And by being good at college, you mean you would just be using AI to do all your work?

I would be using AI to augment my natural human abilities, Kevin.

Something I suggest you try.

Yeah, well, that's, I mean, no, seriously, like, we really see this as a tool for understanding things, right?

Like, you have word processors help you create a document and Photoshop helps you, you know, adjust pixels in an image.

This is a tool that helps you understand things.

And like if you are trying in good faith to understand the material as a student or as a knowledge worker or as a writer or whatever it is, like this is a tool that should really help you do that better.

Yeah.

What's next for you personally?

Are you going to stick around at Google and build some more stuff?

Are you yearning for the solitary life of a writer again?

I am not yearning.

I am really enjoying this.

And there's so much to build.

And to be in the middle, you know, the most important technological change of my life, working with really interesting people on this thing that I've always wanted, like, I would kind of be an idiot to stop now.

Yeah.

Also, the snacks are way better at Google.

Unbelievable, Gavin Deering.

All right, Steven Johnson, thanks so much for coming on.

Thanks, guys.

Wanna come back?

Time to pass the hat for playing a game at Hat GPT.

JP Morgan Payments helps you drive efficiency with automated payments and intelligent algorithms across 200 countries and territories.

That's automation-driven finance.

That's JPMorgan Payments.

JP Morgan Internal Data 2024, Copyright 2025, J.P.

Morgan Chase and Company, All Rights Reserved, JP Morgan Chase Bank, and a member FDIC.

Deposits held in non-U.S.

branches are not FDIC insured.

Non-deposit products are not FDIC insured.

This is not a legal commitment for credit or services.

Availability varies.

Eligibility determined by JP Morgan Chase.

Visit jpmorgan.com slash payments disclosure for details.

At Capella University, learning online doesn't mean learning alone.

You'll get support from people who care about your success, like your enrollment specialist.

who gets to know you and the goals you'd like to achieve.

You'll also get a designated academic coach who's with you throughout your entire program.

Plus, career coaches are available to help you navigate your professional goals.

A different future is closer than you think with Capella University.

Learn more at capella.edu.

All right, Kevin.

Well, I have been so focused on Meta Connect this week that I have not been keeping up with the tech headlines as well as I should have.

And that made me feel like this would be the perfect time to play a game of Hatch GPT.

Yes!

Now, Hatchy PT, of course, is our game where we take headlines from the news, we riff on them a bit, and when one of us gets bored, we say to the other person, hey, stop generating.

Yes.

So, shall we begin?

Let's do it.

Okay, Kevin, this is a big one.

OpenAI CTO Mira Marati just announced that she's leaving the company.

So, as we stepped into the booth today, we saw this one come over the wires.

Mira Marati, of course, the chief technology officer of OpenAI, announced that she was leaving after six and a half years.

She posted posted a note online that said, after much reflection, I have made the difficult decision to leave OpenAI, adding, there's never an ideal time to step away from a place one cherishes, yet this moment feels right.

Now, Kevin, I always feel like in corporatees, the phrase, there's never an ideal time to step away, translated into English means, I know that I'm leaving at a terrible time.

Do you feel that way?

I thought it meant I got fired.

Well, I'm very confident that Mira Mardi did not get fired.

Yeah, I don't think she got fired either, in part because she's not the only person who left this week.

A few hours after Mira made her announcement, two other senior leaders at OpenAI also announced they were leaving.

OpenAI's chief research officer, Bob McGrew, and a vice president of research, Barrett Zoff.

Like Mira, they both sort of gave vague explanations for why they were leaving, like wanting to explore new opportunities and take a break.

So whatever is happening, it seems like a very big deal.

Look, it is.

I mean, I have had people very close to the company Inside Open AI who says, you know, when Sam Altman is out there being the AI diplomat to the entire world, traveling the globe, raising money for his chip projects and everything else, Mira is the one who is overseeing the company's roadmap and really handling all the day-to-day operations.

So, you know, on one hand, this does feel pretty shocking.

But on the other hand, I don't think we can say that we are surprised.

These are just the latest in a wave of of people to leave.

I would not be surprised if Mira started her own AI company.

She says she's stepping away because, quote, I want to create the time and space to do my own exploration.

And whatever that exploration turns out to be, I'm sure Andreess and Horowitz will offer her $100 million for it.

Yeah, I mean, it's just really remarkable how much this company has changed in the past year or two.

There was a photo that sort of became semi-iconic of the sort of leadership of open AI.

I think sitting on a couch together, and it was Sam Altman, Mira Marati, Greg Brockman, and Ilya Sutzgever.

And that was sort of the highest-ranking leadership of this company just a year ago.

And now Ilya Sutzgever has left the company, Mira Marati has left the company, Greg Brockman is out on an extended leave.

So of that group of four, it is just Sam Altman who is there as of now.

What do you think is happening here?

What's the real story?

Well, you know, I was,

my boyfriend is reading a book about the history of Silicon Valley right now, and he just shared this passage with me about the rise of the semiconductor industry.

And one of the first big semiconductor manufacturers was Fairchild.

And it was a very similar situation to OpenAI.

So many people left Fairchild and started their own silicon companies that they were known as the Fairchildren.

And the companies that they created are the reason that we wound up calling this Silicon Valley.

So if you view the rise of AI as kind of a tectonic shift in the technology landscape.

We might be seeing the same thing here, where OpenAI really was one of the biggest incubators for this technology.

And one by one, all of those co-founders said, you know what?

I bet I could do it a little bit better, do it a little more my way and make a little bit more money if I did it somewhere else.

So you think this is about her wanting to do her own thing?

I have some questions here.

Obviously, we'll wait to hear more about if any more reporting comes out about why Mira Marati has left the company.

But I also saw another story today about OpenAI that made me think that it might be related, which is that they are reportedly working to transition to a traditional for-profit company.

Right now, as we know, OpenAI is governed by the board of a nonprofit.

So this transition would basically do away with that.

The nonprofit would be a separate entity and would only hold a minority stake in the sort of conventional for-profit company.

So I just wonder if part of this is just, this is not the kind of project that these people signed up for back when Open AI was founded.

Which gives them all the more reason to leave and start their own thing, right?

Like if those stories do wind up being related, I feel very confident Mira Marati will be working on AI once again very soon.

And as we know, Open AI has said that it no longer punishes whistleblowers by withholding their equity after they leave the company.

So Mira Marati, your invitation to Hard Fork is wide open.

We've got a spot for you next week.

All right.

Stop generating.

This next story is a double backpedal.

And there are two stories on this slip.

One of them is about Elon Musk's ex backing down in Brazil after defying court orders in Brazil for three weeks.

When you say Elon Musk's ex backing down in Brazil, are you referring to Grimes?

No.

Okay.

X the company.

Oh, yeah.

Formerly known as Twitter.

After defying court orders in Brazil for three weeks, X has capitulated.

In a court filing last Friday, X's lawyers said that the company had complied with orders from the Brazilian Supreme Court in hopes that the court would lift a block on its site.

That is a sharp reversal.

It spent weeks insisting that it would not back down, and then it backed down.

Okay, that's story number one.

Story number two, in the same vein, is about Telegram, another company that we've talked about on the show very recently.

Telegram CEO, Durov, says app to provide more data to governments.

The story comes from Bloomberg.

Basically, after insisting that it wouldn't do this, it wouldn't give out data to governments in response to legal requests.

Pavel Durov, the chief executive officer of Telegram, said that Telegram will now provide users' IP addresses and phone numbers to relevant authorities.

Casey, what do you think about these two stories?

Well, you know, first of all, I think they're related.

Second of all,

the Telegram story has made me realize that Pavel Durov and I have something in common, Casey.

Well, we both came to profound realizations while sitting in a French prison.

Did I ever tell you about the time I was in a French prison?

No.

I stole a loaf of bread to feed my family.

I think I saw a play about it.

Yeah, it was a very good musical.

But look, here's the deal.

I think what we've really learned here is that after a long period of countries trying to rein in tech platforms through passing laws and passing regulations, we're now moving into what I like to call the prison phase, where

Supreme Courts and law enforcement officers are going out there and they're saying, you are either going to listen to us and you are going to obey the sovereignty of our country or we're going to throw your executives in jail or we're just going to ban the whole platform.

So this is a real ratcheting up.

And for all of the big talk that Pavel Durov and Elon Musk have done over the past year about how they were in a fight for free speech, when push came to shove, they were not willing to spend the rest of their lives in jail.

All right, stop generating.

Kevin, meet the autocado and a robotic burrito bowl maker at these these Orange County Chipotle locations.

Now, this story is very personal to me because I grew up in Orange County.

And according to the Los Angeles Times, robots that fully flesh out avocados and build burrito bowls are now making food for customers at two Chipotle Mexican grill restaurants in Southern California as automation and artificial intelligence continue to enter the fast food industry and raise fears about the supplanting of workers.

And Kevin, I'm told the Adocado can cut, core, and peel avocados in 26 seconds on average.

How does that compare compare to your avocado technique?

Wow.

I'm pretty fast with an avocado, but I will say I don't hate this because I recently learned that cutting avocados is the cause of many trips to the emergency room.

Is that right?

Many, many is one of the leading reasons that people cut their hands in the kitchen because you know, when people are cutting avocados, sometimes they hold, they'll split the avocado in half, they'll hold a half, and then they'll cut it while it's in their other hand.

And many times, my friend who had injured themselves this way was saying, people will just cut through the skin of the avocado into their hand and have to go to the ER.

Well, I mean, that's why there's that saying, make guacamole, cut a little holy.

And that's how you're supposed to remember to be careful.

But what do you think about

the automatic robot nature of what's happening to Chipotle?

So obviously, I would be worried if I was a Chipotle worker, but I think in some spiritual sense, like Chipotle is already kind of, you know, this like slurry that people just sort of shovel into their mouths like they're horses at a trough.

It's a conveyor belt for beans and rice.

Yes.

So I don't mind this.

I will be interested to see if they can actually automate more parts of the Chipotle process

because this is just one part of the assembly process.

But I have to imagine that Chipotle and many other fast food chains are working on using robots for things like flipping burgers.

Here's my prediction.

If you've ever been to Chipotle, you know one thing, and that is that guac is extra, right?

You know this.

You want to get guacamole on anything, you're going to have to pay up.

I think in the future at Chipotle, humans are going going to be extra, right?

You can either go to the autocado and you can get it cheaper, but hey, you want that burrito made with the love of a human being?

Well, it's going to cost you about another two bucks.

I'm not sure the burritos are made with the love of a human being today.

All right,

stop generating.

Okay,

next up.

Oh, this one's a talker.

This was all over my feeds yesterday.

Marquez Brownlee says, I hear you after fans criticize his new wallpaper app.

So this is from The Verge.

Casey, you'll remember remember former hard fork guest Marquez Brownlee, aka MKBHD, the famous YouTube tech reviewer.

He launched a wallpaper app called Panels this week as part of his iPhone 16 review.

The app was criticized for its price.

A subscription to this app costs $50 a year or $12 a month.

And people were very skeptical that any app for wallpaper for your iPhone could be worth that much money.

And the app also asked to track user activity across other websites and location data.

Marquez sort of backtracked a little bit.

For example, he said they were going to fix the excessive data disclosures.

So, Casey, what did you make of this?

This was a big deal.

You were very mad.

It made me really worried about the upcoming launch of the first official hard fork product, which is a bag of rusty nails.

You know, like looking at how this app was received, I thought, Kevin, we're going to be in for it.

But, you know, look, I really like Marquez.

I have a lot of respect for him.

I think he's like built something really incredible over there.

But, you know, two two things come to mind.

One is when you are a critic and when you are a tough critic, anybody you criticize is going to be out there waiting for you with long knives.

And I think Marquez arguably published the most critical reviews of his entire career this year.

He trashed the humane AI pin, the Rabbit R1 AI gadget, and also an electric car.

And he really sort of built up a reputation for himself as somebody who was really hard on other people's products.

So when you release your product, if it is not absolutely polished and dialed, you're going going to hear it.

The other thing I've been thinking about is just, this is actually an indictment of YouTube, right?

Because even though, you know, YouTube does share a certain amount of revenue with creators, even the most popular creators there, none of them ever just are content with the money they make there.

They feel like in order to be able to live their lives, they have to get into these brand deals.

My friend Eli Patel over at The Verge says, at the end of the day, every YouTube creator has to turn themselves into a miniature advertising agency.

Right.

And so it's pushing people like Marquez.

And it's not like this in other media.

Like Wolf Blitzer is not out there hawking ringtones.

He doesn't have to because he makes enough money from his day job.

And I've got to think that if YouTube was giving people like Marquez, who are among the most successful people on that platform, you know, enough ad revenue to support whatever their plans and ambitions are, they wouldn't have to go do all doing all of these side projects.

Yeah, and now look, you know, maybe some of them would do it anyway.

And, you know, certainly I'm sure some individual creators are just sort of greedy.

And I don't want to totally let creators off the hook who go out and, you know, produce sub-par products.

And it does seem like this was a really sub-par product.

But, you know, to me, the more interesting story about this is like, what is the actual YouTube economy?

And the truth is, it's pretty bad.

You talk to particularly anybody who's running more than a 1% operation on YouTube, how it's going.

And pretty much the best thing that they'll tell you is that it's going about break even.

But there are a lot of people losing money out there on YouTube.

And so when I see something like this, I just think, yeah, this is really kind kind of a YouTube story, too.

It made me think about the other big sort of YouTube brand extension fiasco, which was the Mr.

Beast Burger Empire.

Yeah, you know, I still have indigestion when I'm thinking about the Beast Burger I ate that one time.

So Mr.

Beast, for those of you who don't remember this, he launched a line of burger restaurants that were basically sort of ghost kitchens that would all sort of make these burgers and put Mr.

Beast's branding on them and send them out.

And, you know, many of them got very bad reviews.

There was very poor quality control on these burgers.

And it ended up being a pretty big stain on Mr.

Beast's reputation.

You know, no one thought that he was cooking the burgers himself, but I think this really is the flip side of the parasocial relationship.

Is that a burger joke?

Oh, that's that.

I didn't mean it as one, but now that you mention it.

I think the one thing that people really love about YouTube are these parasocial relationships that fans get to have with their creators where they, you know, they trust them more than you would trust someone on the evening news telling you something because they seem authentic.

Maybe they're filming in their bedrooms or someone that's very personal.

Maybe they're sharing more intimate details from their life.

You end up sort of feeling like you know this person, you trust them.

And so when someone on YouTube sort of leads you astray or recommends a product that maybe doesn't live up to your expectations, it really makes you sort of question them as a whole.

And I think that's what Marquez has stepped into here.

And so I think he'll recover from this.

He's obviously got a long body of work on YouTube.

And I think he will, you know, he will make this right.

But I think it just shows how fragile these parasocial relationships can be.

We've probably already talked about this too long, but I do want to say one more thing about this app because as bad as it is, I do actually want to stand up for the idea of paying people for their art.

Right.

And one thing that I think this app got right was at least they said that they were going to share the proceeds of the app 50-50 with the artists.

And if you are worried that AI is about to put every human artist out of work, an app that comes along and says, we're going to share that money with artists, I think it's at least worth asking yourself whether some of those artists deserve a living wage, right?

And that at the end of the day, 50 bucks isn't that much to pay to, you know, keep somebody's lights on at their house.

Yeah.

Yeah.

Are you a wallpaper guy?

Do you use wallpapers?

I know.

I change it like two or three times a year, but like, you know, is it just a photo of my face?

No.

Yeah, it is.

And I say, I'm going to get them someday.

All right.

Stop shirting.

And now, a next-level moment from ATT Business.

Say you've sent out a gigantic shipment of pillows and they need to be there in time for International Sleep Day.

You've got ATT 5G, so you're fully confident, but the vendor isn't responding, and International Sleep Day is tomorrow.

Luckily, ATT 5G lets you deal with any issues with ease, so the pillows will get delivered and everyone can sleep soundly, especially you.

ATT 5G requires a compatible plan and device, coverage not available everywhere.

Learn more at ATT.com/slash 5G network.

At Capella University, learning online doesn't mean learning alone.

You'll get support from people who care about your success, like your enrollment specialist who gets to know you and the goals you'd like to achieve.

You'll also get a designated academic coach who's with you throughout your entire program.

Plus, career coaches are available to help you navigate your professional goals.

A different future is closer than you think with Capella University.

Learn more at capella.edu.

Hard Fork is produced by Rachel Cohn and Whitney Jones.

We're edited by Jen Poyant.

We're fact-checked by Caitlin Love.

Today's show was engineered by Daniel Ramirez.

Original music by Alicia Baitube, Marion Lozano, Rowan Nemostow, and Dan Powell.

Our audience editor is Nell Galokli.

Video production by Ryan Manning and Chris Schott.

You can watch this whole episode on YouTube at youtube.com/slash hard fork.

Special thanks to Paula Schuman, Hui Wing Tam, Dahlia Haddad, and Jeffrey Miranda.

You can email us at hardfork at nytimes.com.

Maybe take the transcript of this podcast and put it into Notebook LM and see what kind of podcast it makes.

No, don't do that.

What if it's better than us?

Did you know Tide has been upgraded to provide an even better clean and cold water?

Tide is specifically designed to fight any stain you throw at it, even in cold.

Butter?

Yep.

Chocolate ice cream?

Sure thing.

Barbecue sauce?

Tide's got you covered.

You don't need to use warm water.

Additionally, Tide pods let you confidently fight tough stains with new coldzyme technology.

Just remember: if it's gotta be clean, it's gotta be tide.