Big Technology Podcast - Meta Exit Interview — With Mike Schroepfer

Episode Date: June 8, 2022

Mike Schroepfer is Meta's former chief technology officer and currently a senior fellow at the company. He joins Big Technology Podcast for a look back at his time as CTO: What went well? What could've been better? What changes would he have made in retrospect? Listen for an episode about the challenges of enabling massive scale, both from a technical and societal standpoint.

LinkedIn Presents. We are joined today by Mike Shrepfer, a senior fellow and former CTO at Facebook, aka Meta. This is going to be the first time we ever do a show like this, and I'm really excited about it. It's going to be exit interview format. We're going to talk to Shrep about what went right, what went wrong at Facebook, what he can learn from his tenure. And then I also got a bunch of questions from the exit interview handbook that I promise I will ask. So stay tuned. It's going to be fun.

Shrep, welcome to the show. Thanks. Excited to be here. How did you get your nickname? Were people too lazy to say the whole Shrepfer? Because it doesn't sound too difficult for me to pronounce, or is that something that you insisted on? I think, no, I've definitely not insisted. And Michael was like the most popular name for like 20 years.

So I think it was just, it was in college and there were too many Michaels around. So it was just easier and shorter. And my very first job out of college was in a big open floor plan. And I remember being like, okay, I'm a professional now. People should know me as Mike. Shrep doesn't really sound that professional. So everyone in that company knew me as Mike, except when some college friends came to visit in my big open office, like, hey, Shrop, what's up? And everyone's like, oh, Shrep, that works a lot better than Mike.

And so I just gave up. Right. Yeah. And I think that out of everybody at Meta, there's two people who are one named people. There's Zuck and there's Shreep. So it's nice to be on the line with you. So I want to, first of all, get a sense as to how long you were at Meta for our listeners, if you can share that. And then how did the company change over your time there?

Well, I joined in, it was late summer, fall of 2008. And so to put it in context, at that time, MySpace had more users than Facebook. So MySpace was the big social network on the block. And Facebook was the new up-and-comer. And so it's been about, you know, coming up on about 14. years since then. And so obviously, we have changed a lot over the years. It was a much smaller company, much smaller website at the time, you know, that was just around the launch of the mobile app store. And so mobile was a distance away. And so it was, it has changed a lot in those

13, 14 years. And what position did you come in as? So I joined and I was, I was, actually my very first job was director of engineering and for about two or three weeks I ran part of engineering and actually part of product management and design. And then I had a pre-planned vacation that I had planned like a year in advance. So I took off for a while and came back. And when I came back, Mark, you know, sat Chris and I down and said, Chris Cox. Chris Cox was running HR at the time. And it was like this sort of half engineering and Chris is in HR. It just makes sense. Like how about Chris, you run product and Shrep, you run engineering? And we're like that makes a lot more sense. And so we swizzled around and Chris became the head of product and I

became the head of engineering. And this was in early or fall of 2008. And that's how we operated for quite a long time. I'm going to get to some of the what went well, what went wrong questions. But I, you know, you mentioned that it was just around the time of mobile getting off the ground. I remember I was at age, maybe it was 2010 or 2011 where I wrote the story that for the first time ever, more media was consumed on mobile devices than it was on desktop and it was all of a sudden after everybody was yelling a year of mobile probably since the moment you began it had come and it came fast and it didn't really lead anyone to have any time to spend a while making that transition. Facebook had a very interesting moment there where if I'm not mistaken

and Mark Zuckerberg was very intent on a mode of developing where you would build hybrid apps and then moved to a mode of building native only for Apple and Android devices. So I'm curious from your seat what that was like making the transition to mobile inside Facebook. Well, it's, I mean, it's important to back up because it did occur very fast, but there were many years. When I first got there, mobile wasn't on the horizon in 2008. I mean, I remember when I was thinking about joining Facebook, I had friends try to convince me not to join because the issue actually was, could you build a business in a social network? And the word on the street in 2008 was no, you know, that MySpace was having a lot of trouble.

If you remember, they had signed this big deal with Google that had minimums that they were going to meet and they couldn't meet them. And then people had a long list of companies before, Friendsster, you know, AOL instant messenger. and the word on the street was like, well, lots of people use these things, but you can't actually build an interesting company business out of it. And so really the first job was, A, just keep the site scaling because Friendster had blown up because it couldn't actually keep up the load. It kept crashing. And then B, assuming we could actually scale it to the demand of users, how do you build

a business on this? And that was sort of building the ads business on the web, which started as the right-hand column. So you had kind of your nav on the left. and your fee in the middle and these ads on the third column on the right. And we had kind of just gotten that working. And that was like the, can this company survive moment?

And then all of a sudden, you know, it's like 2011-ish time frame 12. You know, we saw the user base shift very dramatically from web to mobile. And so you had a double problem here. You know, the problem number one was we had built everything on the web using PHP, JavaScript, web technologies. And now you move to mobile, and it's a whole different technology stack. So you need Objective C.

You're building an iOS on Android. You're building a Java. So like the language that our programmers work in is now totally different. All the tooling is different. And then you have this like giant screen going to a tiny screen. So where we had three columns on the web, you have one column on mobile. And we had built our business on the third column. So it's kind of like, what do you do with that?

third column. And so we basically had these two challenges at once, which is technologically, how do we build on mobile? And then business-wise, how do we actually make our business work on mobile? We kind of had to tackle both at the same time. So, you know, when we looked around, I said, look, we've got hundreds of engineers who are trained up and best in the world at building web technologies. We've got to either cross-train all of them into a brand new technology stack on iOS, or if we can bring some of their knowledge and expertise over onto mobile, that will allow us to move much faster. Said much more simply, like you have a whole bunch of people writing one version of your

product here in this form, and then on mobile, you had to basically like copy the whole thing and do it again in a totally different language. And by the way, do it again a third time if you want to also do it on Android since it's totally separate. And that just felt untenable at the time. So we tried to figure out how much basically code and knowledge. could we share between web development and mobile? And that was, you know, our first sort of attempt to scale up on mobile. And as you say, it sounded good at the time, but it was quite

challenging technically to build a product that was what people expected, which was high performance. When you scroll, you know, a feed on your phone, you don't expect it to stutter and jutter and a loading wheel. You just expect kind of buttery smooth, you know, feed. And that was quite challenging to do at that time. Hybrid. Yeah. That technology stack. Those apps did not work very well. Yeah. No, we had to basically reverse course. And we built a new version. We started on iOS, and we basically started from scratch and said,

rewrite the whole thing, start from scratch. And your number one feature is performance. And everything else kind of can fall off the boat, an order. order to make the performance target. And this is, you know, all seems great now because it all worked out. But at the time, you know, we had tons of business objectives, tons of ideas, and we basically had no new features on the existing app, because we still had the existing app in the app store, still a product that hundreds of millions of people used. And we said, stop development on that except for security and major bug fixes.

And everybody just wait nine months-ish until this new version comes out. And you don't get to deploy any new features. And being able to sort of silo and focus that team and build it who built a sort of native first product was a, was a huge challenge at the time. But when it came out, people, people love the new version. And this is the 5.0 version of the iOS app. And it was a, you know, built a huge success from there. Right. And it's interesting that you started talking about with how can you build a business for Facebook.

And this is a story, the stat that, you know, I like to cite whenever I talk to people about this is that. Now, Facebook's revenue is 95 to 98% mobile or something like that, somewhere in that range. Yeah. And this is the sort of platform transition that typically kills companies. So we built a business on the web, different technology stack, different product experience, move to mobile. That's the time when usually most companies hold on too tight to their existing business because it's getting disrupted. And it's hard to remember at the time. Most people say, like, ah, people aren't going to buy stuff on their phone.

like, who wants to type their credit card into this little device? Like, eh, that's not going to happen. And so it wasn't even just like it's hard. It's like people didn't believe it would exist. And so my experience that meta has been like every three years, there's some what feels like an existential crisis. When I got there was like, can we scale the site? Can we build a business?

And it's like, cool. We're kind of getting through that. It's like, oh, you got to do it all over again on a whole new platform and like do it fast. And this is, you know, part of the, honestly, the power of a founder or CEO and is, you alluded to this, as Mark at the time, basically just through the switch all the way to mobile and said, we've got to do mobile. It's the future. And not only we're able to silo this team and like stop development, which is a big deal for a company, he said, look, anytime you come in and bring me a preview of a new product, new experience, it has to be on mobile. Like, I don't want to see anything that's a web mockup. And people didn't think he was serious. And I remember the first team that rolled in is like, here's our new thing. And like, here's our new thing. And like, And he's like, wait, these are web locks. I'm like, yeah. Like, come back with mobile mocks.

I don't want to do this review. And they're like, oh, you're serious? So, like, that sort of thing, I think, is needed in these transitions. Does Facebook go a debt of gratitude to Android and Apple for developing mobile operating systems that could take people from that, oh, no one will ever transact on phones to actually, hey, wait a second, advertising is legitimate business on phones? I mean, I think it's fair to say that Facebook obviously wouldn't exist without the internet first, and a lot of technology is built on there, and then the explosion of consumers on mobile devices. There are more people with mobile phones than would ever have desktop computers. So I think it's a huge part of the growth story. Part of why we're so interested in building the next platform, ARVR and the Metaverse, is because I think these platform shifts can be big danger for companies.

but for company like Meta, Facebook, I think it's a huge opportunity for us in the future. Okay, but so that sounds like a yes in terms of some form of, you know, debt to these companies for creating the conditions that you were able to do business on their operating systems. I think like having mobile operating systems out there and ability to deploy our apps has been really great for Facebook. Yeah. Yeah. Okay. Let's get to some of these. That was, I'm glad, again, I'm glad we did that diversion. So here, I wrote some exit interview questions. Maybe that our discussion just now leads into the first one, which is, you know, you go from being smaller than MySpace to being 3.6 billion people using your products every month.

What went well? Well, I mean, I think the fact that it worked, like there's a huge survivor bias in everyone. It's just like you assume because this thing exists, it was easy along the way or even that it was bound to happen. And I think this is this like techno determinism that these technologies will exist or these companies exist is just not my personal experience. It was, you know, one fire after the other along the way. And I think that, you know, looking back at the time, this is a heck of an answer for the what went well part. Sorry, go ahead. Well, no, I just like, what went well is like we survived is like part of it.

And like back of the day, there wasn't a software stack for us to build on. Like people had built, you know, web search. They had built new sites where. you could serve the same article to billions of people, no problem. You just copy it, easy, peasy. At the time, we were like trying to build this thing where anyone could write on anything. They could comment or like. And then anyone could view that thing. And they expected to get the most recent version at any point in time. And so at the time, it was, I remember Justin, it's like, oh, gosh, is Justin Bieber going to post today? Because Justin Bieber posts,

and all of a sudden, millions of people want to like and comment. And, like, the whole world expects that to just, like, work. And there wasn't, like, a product I could, buy on the internet or go to fries and buy, like we had to build a whole new software stack to do this that worked and worked at like massive scale. And then we had to kind of build the whole team along the way that built this. And then while we were building this new software stack to scale, we also sort of made this decision, said, look, consumer like preferences change. We have to have this culture of speed, our ability to deploy code. So we actually accelerated our ability to deploy code over time. When I first got there, it was a weekly push. So we would

change this site in major ways once a week. It's now a continuous push, which means literally as we're doing this podcast, like changes are happening. And so, and everyone I asked for advice on, I was like, no, no, no, you go slower as you get bigger because it's harder. And we said, no, we thought ability to move quickly was important. So what we needed to do is build really robust tools and really robust process to make this all work. And so I think that that, you know, a lot of that is what right. You know, the other thing I'd say is, um, a long the way, we built a lot of tools for ourselves that have become a sort of industry standard for how people build stuff. Like, if you talk to someone's like, they're going to build a web app,

they're going to build a mobile app. They're going to use React probably or React Native on mobile. And these are how we build our own products. And if you're doing state of the art AI research right now, you're, you're, you know, three out of four times going to be using PyTorch, which is a tool, again, we built for ourselves, you know, and if you're going to do storage, you're going to there's rocks to be. If you're running PHP, you're going to use HHVM. These are all products that we built and have opened sourced and have become sort of part of the DNA of how people build things on web, on mobile, and in AI. So it's not just one technological realm. It's sort of several. So that ability to sort of scale, solve problems and then do it in a way that like people

kind of love as a developer is, I think one of the things that went really, really well. Why open source those code bases? There's lots of reasons. I mean, I think that, you know, one is just leverage. This is basically just for definitions. You're giving this stuff away to developers that could potentially compete with you. Yeah, everyone can use the same tools we're using. And we actually did this in our hardware too.

So most people don't know about open compute, but it's basically like go use the designs for our servers and data centers yourself. But, you know, when you talk about things like an AI research framework or a web development framework, framework, our mobile development framework, that's like need tons and tons of people have. And so I'd rather people collaborate together to build the best one, rather than everyone build their own special version that's not as good. And I also like, my favorite part about it is it keeps us honest because our culture inside the company is one of developer pull rather than centralized push, meaning you don't have a mandate in the company that you must use this technology, which happens in big companies.

And what happens is that technology sort of was really good and then sort of isn't as good as what everyone else could use. And so you're fundamentally slower than every other company out there. And by using open source, like everyone else who uses these products outside the company doesn't have to. So it's got to be good. And if people stop using it, that's a really important signal to us that there's something wrong with it that we need to fix. Or we should switch to a different technology. Because like my job is to always have our teams on the absolute latest, greatest, best technology. So they're as productive or more productive than anyone else in the industry.

And being an open source on the core of what we build is a really good way to ensure that. And so it's why we did it way early day in the early days at HHVM, then we did it with React. And then when AI has been the big revolution, it's by torch. And again, back to what you're saying, it's, I think the thing I'm most proud of is us ability to jump from platform and technological realm, web, mobile, AI, ARVR. that is pretty unusual. Usually have a company gets really good in one thing, and they sort of miss the motion. So it's actually our ability to migrate these things and take our approach to them that I'm most proud of. And okay, what went wrong?

What went wrong? Well, there was lots of challenges along the way. I think we've seen over the last five plus years, you know, the sort of struggles with content moderation and, you know, how to manage that. I think the world is still sort of sorting through this. And not only what the right rules of the road are in terms of what policies people want to balance sort of safety versus expression, but then how to implement them and how to build them in an operational robust way so they scale to billions of users.

I think that's been a real challenge. And I think one of the lessons we're trying to take to the Metaverse is to think a lot further in advance of all of those things. So what are the downsides? What are the possible ways bad actors could abuse the system? And if I knew that that was going to happen, what are all the tools and features I can build in my products now rather than wait for it to be a big issue? And I think that's been a big lesson for us over the last, you know, many years. A few people have built a product or helped, you know, build the infrastructure to enable a product to scale in the the way you have. I'm curious if, you know, looking at the way that that Facebook in particular

scaled, you know, over time and having built the infrastructure to enable it to scale, sometimes, you know, critics would say too fast, do you have any reflections on, you know, on products becoming that big that quickly? Well, look, I mean, I think that what attracted me to Facebook in the early days was, you know, the basics of what the company, was trying to do was filling a fundamental human need, which is like, do people want to stay in touch with other people they care about around the world with as little friction as possible? And I think that's an emphatic yes. And I think that that is definitely a net good to society in terms of people being able to keep in touch across the world. And I think there's

some challenges that come along with it. And that's what we just talked about and what we're learning about. But it's hard for me to sort of run the reverse and say, like, actually, let's go back to a time when, like, messaging someone costs you 10 cents a message. Like, I don't think that that's a better world for people. And so I think there's a question of, you know, how much of this, I think when anyone's building in a high growth industry, how much of this is truly unknown versus if you just sat and think about it, say, let's spend the time to Red Team this, a war game at, let's think about all the ways in which people might use this for ill and let's let's like plan out our defenses up front. And I think that's the big,

you know, lesson for me and how we're building, you know, four products in the future. Yeah. And just one last question about growth. I think sometimes people will look at growth and they say as long as that line is going up and to the right, then life is good. Things are going well for the company. What would you say to people who look at that as the sole metric? I mean, I think a single metric is always dangerous and never how we've operated the company because there's so many ways in which that can go wrong. Among them can be, you can be on a really near-term local maximum where it's going up, but it's going to stop and it's going to come back down the other side. There's been plenty of businesses that look like that, plenty of fads that occur in the industry. And so you

always got to be clicking one level down and looking at the fundamentals. Like, are people happy, are they enjoying it? Do they want to come back over the long term? Because if you're building a multi-decade business, that's what matters, not like, his next quarter could. And then, you know, all the other things we talked about in terms of understanding the other consequences of the business you're building. So I think that the sort of anyone who ever operates, who's operated anything knows that a single metric is a great way to really get yourself in trouble and not in how we've operated business. Yeah. I want to put one finer point on it because I feel like maybe this is a better way to ask the question because I feel like this is what people have said in the past.

Do you think the company grew faster than it could handle? I think it's really hard to run the like historical AB comparison so like I think that you know I don't know so I think that because you'd have to ask what would have grown in instead what would people be using instead and would those things be better or worse it's it's really hard to know I think all I know is what I have learned and what I can take forward now which is you know what I said just thinking hard about you know about what we could know and how we can prepare for those things right now yeah another question that I wanted to ask you is looking at I always look at the reports that Facebook puts out on takedowns, and oftentimes it's like, you know, millions of terrorism posts, hundreds of millions of, you know, pieces of child pornography.

We all know, you know, what some of the content moderators have said about, like, the effects some of this stuff has on their mental health. I'm curious what you've learned about the nature of humanity. I mean, seeing how much bad people have the ability to produce. Has it changed your perspective at all? I mean, look, I'm probably at heart a massive optimist, but I would say that my optimism armor has been severely dented over the last five years. It is hard to not have it when you experience some of these things firsthand and actually look at the content and look at what people do

and realize that sometimes some people are capable of great evil and harm. And you have to understand that to be real about it, to be willing to fight and fight those things and say, these things exist. We can't just plan for the happy path and assume everyone's going to be nice all the time because they won't. It doesn't mean most people are or that I'm despondent about humanity, but it does mean that a loud minority can be really terrible. And it's our job to fight against them and make it good for the average person on the product, which I think is achievable. And I think, you know, when you look at content moderation, for example, you know, on on the platform, you know, I know, I know this is probably not going to resonate for people, but the numbers tell the story. If you look at our, you know, quarterly reports where we lay this out, what you've seen is a steady decline in the occurrence of these things on our product. What we say prevalence, how likely are you to, you know, see one of these things, you know, and in some cases like hate speech, it's gone down by 5x in 18 months, you know, and it's from point one to 0.0.0.2.

2%. You know, and I'd ask how many people listening this have seen nudity on Facebook in the last, you know, three, six months. And, you know, this is a product where there's a big button on your phone where you can just say upload image and you can upload whatever you want. And it could be adult nudity, but that's against our policies. And it's up to us to find that before you do and get rid of it. And six, seven, eight years ago, we would have to do that by someone seeing it reporting it. And now 95 to 99% of the time, we catch it first and pull it down before anyone's seen it. And that, I think, you know, that gives me some hope that we can sort of build the right sort of guardrails around these systems to allow people free expression, not restrict

what you want to say, but like remove the worst of the worst so that people can have a safe and enjoyable experience. Right. And all the nature of humanity question in particular, like I have tended to think that the nature of humans is good. However, like, if I think if I was sitting in your chair where I saw how many awful images are uploaded, and I'm not saying that people see them. You're basically, you've been behind some of the AI systems that catch this stuff before it happens, the fact that Facebook blocks nudity, largely due to AI systems that you've built to detect it before it ever can be posted.

So I'm curious if you, maybe you don't want to answer this one, but what, Your view on the nature of humanity good or something else. Look, I won't lie and I say that there's been mornings that my faith in humanity have been tested. I think particularly acts of violence and hate, you know, the Christchurch shooting, you know, the recent shooting and Vivaldi, like, it's really hard to wrap my brain around the pain people cause. and choose to cause. But I think that if you sort of zoom out and look at the numbers

and say that the vast, vast majority of people don't do this and are good and help their neighbor across the street. And like our job is to bend the arc of society positively. And if you zoom in history, like despite all of the terrible things happening, we're in the least violent time in humanity. And a lot of that is because of progress

and because of opportunity and generating wealth. and technology. And so, you know, this is where it gets me, you know, there are times that I am on the floor and I got to pick myself back up. The reason I get back into the chair and get back at it is I say, what we can do is make tomorrow better than today. I can't, I can't fix today or yesterday. But if we could make people more prosperous, give them hope, give them ability to connect with other people, then I know that that makes tomorrow better. And, and when we open those opportunities, I think people are wonderful.

Interesting. I'm glad we spoke about this. Let's talk about open culture. Facebook has an extremely open culture. You've been part of the leadership. I'm using Facebook and meta interchangeably. Please forgive me. You've been part of the leadership for a long time. And when the documents that Francis Hogan leaked came out, I actually saw your name in a lot of the conversations. I'm curious. And afterwards, Facebook did take some of the permission. that there used to be some of the openness and close it up a little bit. I'm curious if what your view is on the culture and the openness perspective in particular, was that just a little too optimistic, or are you still a believer in the open culture? Look, I still believe that you want everyone involved in an endeavor to have, I mean, the source of the open culture was very practical. Sort of like, if you are missing some piece of information that would help you do your job better, that's like on us, right?

Like, my job was to make sure that every engineer in the company had the best tools, were working in the right direction, and had all the information they needed to make the right decisions. And then I got out of the way. Like all the best stuff we built was not me showing up and saying, I have this great idea. It was creating the conditions for those ideas to emerge. And then in some cases, being an editor and curating and saying,

that's a really good idea. I'm going to accelerate that or protect it. And so I think that requires information. I think on the flip side, you know, there's a reason why people don't live stream their living room conversations, right? If you know the entire world is watching, you will think a lot before you say things, right? And I think when you're having a knock down, drag out conversation about a policy or a product or a thing, you kind of want people to say, here's what I think. I think this product's dumb. I think we shouldn't do this.

I think this is the wrong strategy. And if you're like, oh, geez, this is going up on YouTube, you may be quiet and not express important things. And so I do think that like shining a, you know, spotlight on these conversations causes people to claim up. And so I think that that is a challenge is like we want to get all the information we can to people, want to distribute be as open as possible, but we want to create an environment where people are comfortable expressing their views, their opinions, and add it to the

conversation. And that requires some guardrails to say, But like, look, you're not, you know, your comments aren't going to be live tweeted on the internet. And so you can be wrong. You can use inflammatory language, you know, about how terrible this product is, even though that's probably overstated from how you really feel, but your emotions are catching on. Like, I think that people miss that. And so I think what we're trying to do is say, look, look, we want the information.

We've got to create safe spaces for people to have really direct, honest conversations. So trying to put a finger on what exactly your view is on this. So yes to open culture, but I think you start with openness and say, like, we want to get as much things out to as many people as possible, except when the sort of the largeness of that, you know, the extreme of openness is like live broadcast on the internet, right? Right. And when that starts to cause people to not provide critical information or not express their true views, then you need to create safer contained spaces. And it's it's never been one or the other. Even back in the open culture days, it's not like every team meeting would be broadcast to the entire company. You could have a short team meeting. It's like, oh, we're really like not sure about, is it react? Is it this? Is it face web? Is it native? I mean, when we chose, we went through three different technology stacks for the mobile app that we talked about earlier. We had three teams out there proposed three different ideas. And we had a closed meeting to decide which one to use. We didn't invite the whole company like in a stadium to like watch that decision happened because that would have really clapped.

And I wanted everyone to critique everyone else's approach. It was like a dozen people in a room, right? So I think this idea that open means that there's like a stadium for everyone to watch everything, it like misunderstands how that affects people's ability to have an honest conversation. Once we had that conversation said, this is what we're doing. share that on the rooftops, explain why, talk about the pros and cons, like don't sugarcoat it. So I think there's this balance of allowing people to have safe conversations.

And then once you've done that, like spread information far and wide and making sure that everyone can be involved. Interesting. It's interesting that you said back in the open culture days. So obviously things have shifted after how good. Well, I mean, I think as we've discussed, it's harder when at our, when stuff, stuff is, you know, in this mode. You have to adjust, you know, what conversations happen. But I'm just saying that, like, it's not like everything was, you know, open to begin with, live streamed to the whole company in 2008. A lot was open. Now it's closed. Yeah. But, but it's hard to measure these

things because, like, the company was also a thousand people. Now it's, you know, tens of thousands of people. So, and even back then, we, we had these, you know, 10 person conversations about critical issues. So I think it's like, it's always just a balance of, like I said, I think you start with like get everyone in the information they need, but don't, but make sure you can have a critical conversation where people feel ability to be open. And those two things pull against each other. And I think the question is like at any one instant in time, how to, how to perfectly balance them. Like a lot of things in life, it's balancing equities. There's no like magic right solution to these things. Yeah. Every exit annual, every exit interview manual that I've read has

said you really got to try to figure out why the person's leaving. And you're still a senior fellow at the company but no longer chief technology officer. What's the story? I didn't know there manuals out there. Can you send something to me? I mean, just Google, exit interview questions on the internet. And there are some wild questions out there. Some of them I knew I wasn't going to get a good answer from you. So I didn't ask them. So, oh, okay. Well, thanks for doing your homework. But this is one I think that everyone has. And yeah, I would like to hear about it. I mean, it's pretty simple. It's basically, look, I've been at the company coming up on 14 years, which is a long run by any tech texture.

I've been in tech for 25 years. So, you know, and seeing, started my own company, seeing a lot of different things, things happen. You know, the last few years, I've built a particular passion in putting my own personal time and energy into fighting climate change. I started doing this philanthropically kind of years ago. We have a bunch of stuff cooking there.

And I just, I really was a tale of two loves, which is, look, I want to invest more of my time and energy in this work on climate change. And I feel like if I can put, more days of the week in there, I can do more. But there's a bunch of things that Meta is doing that I really love as well and think are fundamental to the prosperity advancement of humanity, particularly AI and ARVR and a metaverse work.

And so it really was this question of like, how do we, how do I do both of these things at once? And this is why, I mean, people are surprised when I say I'm a senior fellow. They think of it as a, you know, perfunctory title. I'm at the company two days a week. You know, you know, it's a, I'm doing serious work there and able to continue to work in AI, and it's freed up some time for me to work on my work on climate. Mike Trappfer is with us. He is a senior fellow at META, formerly the chief technology officer.

We're talking about a reflection of why he decided to leave and everything he saw at the company, and he just teed up an amazing second half where we'll talk about AI, VR, AR, and I'm going to toss in some climate change. We'll be back right after this. Hey, everyone. Let me tell you about the Hustle Daily Show, a podcast filled with business, tech news, and original stories to keep you in the loop on what's trending. More than 2 million professionals read The Hustle's daily email for its irreverent and informative takes on business and tech news. Now, they have a daily podcast called The Hustle Daily Show, where their team of writers

break down the biggest business headlines in 15 minutes or less and explain why you should care about them. So, search for The Hustle Daily Show and your favorite podcast app, like the one you're using right now. And we're back here on Big Technology Podcast with Mark Shepfer, senior fellow at Facebook. Oh, sorry, META, did it again. And formerly the chief technology officer there. Actually, let's start with the climate change question. During COVID myself, and I think a lot of people who I spoke with said,

if we can't get people to adopt vaccines and protect themselves, or we can't get people to stay inside and protect others, and we have an immediate threat in front of us, we've really lost faith that we can actually do anything toward the climate change problem. I find it interesting that climate change is your next big thing that you want to tackle. I'm sure you might have had some of the similar thoughts. So why are you optimistic that we can actually solve the problem?

Yeah, I think if you come at it from asking overloaded people to do one more thing, you're going to get pretty down about the whole thing. But when you actually look at it and say, wait a second, what are the technological advances we can make that make this an easy yes for people? It's just strictly better. for example, you know, we're now at the point where solar plus batteries is cheaper than coal. And whatever you think about climate change, nobody likes air pollution. You put a coal power plant and a bunch of people are going to get sick and die.

And you replace that coal power plant with solar and batteries. People don't die. It's good for their health. And it's cheaper. And by the way, it also helps with climate change. So, like, there are a lot of those wins on the board yet to be had in a lot of interesting industries. if you've driven an electric car and you realize how little service it takes, there's no oil change, there's no spark plugs, no timing belt, no transmission.

I mean, there's just like a ton of stuff. You just don't have to do anymore. It's basically a vacuum cleaner on wheels. And the cost of operating is so much lower, especially in today's crazy gas prices. You say, okay, if we can bring those to market at reasonable costs, that's good. And you just sort of say, what are all the advances in materials in computing and AI and others that are going to bring things that people like, nah, I want that because that's better. and oh yeah it happens to help with climate change but like also it's better so that's the opportunity

ahead of us and this is why I'm personally so excited about it because so much of what I've learned is how to build and scale technological operations how to do R&D and new areas like AI and AR and VR and that's what I want to you know help accelerate in this industry and I've always felt that well not always actually in particular after COVID that climate change was going to need a technological solution versus like a mass conservation solution you know, after we use technology via the vaccines to actually help get us out of COVID versus anything else. It sounds right to you, it sounds like. Well, I mean, I think, you know, people have done readings in history. My reading of history is like technology is, is the great

accelerator of prosperity because it increases productivity and it removes like tradeoffs. And that's one of the few things out there. And I talked about solar. I mean, LED lighting is another one, you know, 1,000x reduction in cost over the last 20 years, which is just crazy genome sequencing, you know, is just fallen, you know, these curves, you just don't see these curves in nature that look like these massive down to the right, but you see them in technology. And when you can get on one of these curves, it just like throws off all sorts of benefits to humanity. And that's, that's what's exciting about this space, especially batteries, electrification. There's so much to do. I think it will make us a like less polluted, better, happier, more prosperous, productive

world that also will happen to help the climate change problem. That would be neat. Okay. So let's round this out talking about AI, AI, VR and AR. What's the state of AI right now? Because, you know, I think a few years ago when Facebook was working on, you know, M, which was this AI, you know, AI training messaging bot, which I wrote about while I was at BuzzFeed, you know, you built Facebook AI research with Jan Lacoon, who we've had on the show. There was a moment maybe in 2017 where it seemed like AI was rocketing in terms of its ability to do things. and then the buzz died down. It was like when people were putting blockchain in the name of their company to, you know, increase their valuation, you looked at earnings reports and people were saying machine learning in the same way just a couple years ago. And it looks like people have moved on from that fact.

Now, obviously, AI is still continuing to move. I'm curious what you think has been happening as people have been more quiet about it. And, you know, for those listening, Shrepp just gave me a face from across the screen. So I think he's going to dispute my characterization. of the buzz dying down well i think certainly like the hype train and sort of throwing crazy money at it like any new technology you unfortunately have some over exuberance you know and you can debate whether it's a you know a boom or a bubble and we certainly have had that in an i but if you sort of zoom out a little bit and even just take a very short term like 10 year look at

a i and you know back in 2012 is is around the time when the first deep learning network won this thing called ImageNet, which is a challenge Fay-Fa-Lee now at Stanford created to help people build AI systems that can recognize objects and images. And she built this in the late Otts. And for years, it was like this backwater project that nobody cared about. And then all of a sudden, AlexNet showed up. And it was 10% better than the next entrant because it used deep learning. And that was 2012. And then it took us a while. We talked about content moderation. It took us years to get this technology to be useful enough in productive scale in this sort of supervised training mode where you kind of feed the AI, lots of handcrafted sort of data, you kind of spoon-feeded

data from a human. And that sort of what got us through years. And if you again, go back to these standards reports and look at our advancements, it's all from this for the early years. And then in the last few years, we've actually had a second acceleration, which was, you know, I remember we were just like getting this stuff to work. And Jan Lacoon, who would hire a couple years later, it's like, yeah, the supervised learning thing is, it's not going to scale. So we really got to figure out how to make unsupervised learning work where like the machines just like figure stuff out on their own. And I remember like, I was like, are you kidding me? We're like, just got this stuff to work. And you're telling me it's like, it's like the mobile

thing all over again. It's like, oh my God, it just got this to work. Oh, nope, change the platform. And here we are on AI. He was right. And we've gotten unsupervised or self-supervised learning to work across the industry where you build these massive language models that basically can build kind of training sets on their own just by looking at the data and doing tasks like given a sentence, predict the next word, which kind of gives you the test and the answer all in one by just binding text to look out there. And we've seen this massive sort of improvement in the capability of AI systems, even the last few years. We've deployed them internally. And we've built models. This is the other thing that like people are probably going to get bored up, but I just got

to say it. It's just like a couple years ago, we're like, hey, can we build a model that understands a concept in more than one language? So this idea. of hate speech, can we build, instead of building an English one and then a French one and then a, you know, a Spanish one, can we just build one model across multiple languages that understands this concept? And for a long time, the team building this multimodal model was like getting worse results than the best of the monolingual model, except one day they didn't. And they're like, huh, it turns out there's a lot of data in English, and the more you train in English, the better you get in smaller languages. And that gap is just increased over time. And so we now have

these models that are not just multilingual. They're multimodal, so they understand text, images, video, audio, and then they're multi-task. So they're not just like hate speech. It could be bullying, hate speech, a variety of other things. So you have these models that just understand more and more and more over time. And they're getting better and better. So I think that, and it's hard for people to see because most of what AI does is that like, it's like we went into the engine and like souped it up.

It's like I just showed up with like a hundred more horsepower every year and like this magic went into it. It's kind of the way AI is manifesting itself. in the world. It's not so obvious to people yet, but it will be more and more obvious over time as this technology gets deployed in ways that people can see. People are asking, when's it going to do stuff for me? Like I set it out on tasks and it completes them. Is that 10 years away? I don't ever give the 10-year-ish directions. I mean, I think, I don't know. I think that, you know. Well, you guys tried to do it with M and eventually gave up on the project,

the chatbot. It was too early. So how far away are we? I think years, not decades. Okay. Interesting. Did you, by the way, did you see, have you seen this Dolly stuff that Open AI is doing where it's drawing pictures? Yeah. That stuff really blows me away.

Yeah. I mean, the idea that you could, you know, type, you know, pandas playing, you know, tennis on the moon and it produces an image that looks reasonable is, is sort of mind-blowing to people. And I think we're going to continue to see these advances. I mean, we've seen the scale out, too, where you just, you kind of build bigger models and you scale them. And we haven't yet seen like the limit to what can happen there. And there's more that needs to be figured out. But I think there's, it's going to be exciting a couple of years as we advance. I, I, the reason I gave you that frowny face is like, despite all the hype,

I expected things to slow down more than they have in terms of actual progress. Right. So people have made actual like useful production ready progress in these, these models faster than I thought for a longer period of time. Do you think we're going to get within our lifetimes? Oh, this is already in a question that's going to annoy you, but I'm going to ask it anyway, to artificial intelligence that can mirror or surpass human intelligence. I mean, I think this is a, it's a tough benchmark to talk about because humans embody so many different things. And there's, you know, there's so many different ways to look at the delta between AIs and now. I mean, the human brain uses like 20 watts. Most A.I. systems are using hundreds, thousands, megawatts. So we're like, we're missing some

algorithmic tricks for sure. And even if you look at the biggest AI system, out there and you try to do some like rough comparison to synapses in the human brain they're not perfect you know parameters of synapses but you're still in the like less than a percent of the scale so like we're way off in terms of getting anything that looks like or computes like the human brain i do think what we're going to see though we already see this where specialized systems can work better than humans it's like can you identify objectionable content as good or better than a like well-trained person? Yeah, we're getting there. You know, can you translate in hundreds of languages? Yeah. Can you read x-rays, you know, and assist humans? And the other thing is, I think that the best

systems are actually combinations, because computers are good at things that humans aren't and vice versa. And so people always are obsessed with this replacement. But like, for, in x-rays, for example, I think you're going to see computer-assisted radiologists. And they'll be the best out there, because the humans will add to it. And they'll be better with a computer, but you still want the human in the loop. So I think we're going to see a lot of those things occur over time. I think you're going to see self-driving, you know, as much as that's been, you know, sad for everyone. We're making real progress. There are level, you know, level four systems out there in the world. And they, you know, can perform really well in certain circumstances. And so I think people

will be surprised how quickly these things show up, you know, in the coming years. Do you trust a Tesla on full self-driving mode? That's a tough question. I have a Tesla and I do not run it in self-driving mode. Okay. Interesting. AI, no, sorry, AR, Google tried it with glass. The problem wasn't technology. The problem was that it was awkward where people felt that there was a spy cam on people's faces at all times. Now we're in a mode where everyone wants to do AR again.

Why do you think it will be different this time? I mean, Apple's out there trying to do their own device. People think, oh, it'll be stylish. it will solve all the problems. Again, it wasn't the technology. What do you think? I disagree. I mean, did you use the Google Glass?

I did. I think it was just too early. Like, I think they tried really hard and they built something great using the technology at the time. But if you remember the display, like, I had to go like this to look at it. And it was a tiny postage stamp little image there in the corner. And it was like, what does this do for me? There wasn't a good answer.

So I think the consumer utility was extraordinarily low. And then the sort of the side effect factor was really high, as you say. And I think we're just not there yet. But if I told you, look, you know, these glasses that I'm wearing now, they allow me to live caption the world. So if I'm in a noisy room, I can get like subtitles. And oh, by the way, it'll live translate. So if I'm traveling around the world, someone can be speaking at me and I can get a live

translation. People are going to like that. They're going to want to use that. And the more that the AI develops, the more as I'm walking around, I'm like, oh, where did I put, you know, I got a beep, beep, my phone. It's like, you say like, hey, where's my phone? You left it on your desk and here's a photo of it. Like, people are going to like that.

So I think we haven't delivered the utility sort of compromise tradeoff to like get product market fit yet for AR. That's why you don't see those products even today in 2012 years and years after Google Glass because they're still too bulky and the AI needed to provide the real awesome experiences isn't there yet. But we're making advances. Like I always look at the slope, not not the intercept. So like where are we making progress?

And we're making progress along all dimensions. lighter, better displays that are brighter and wider field of view, AI technology that will deliver amazing experiences. I don't exactly know when they'll intersect, but they will eventually intersect. And I think that it will be amazing. On VR, I'm of the belief that outside of gaming, VR is going to be an enterprise tool for a long time. I think Facebook has done this, build some of the rooms where you can speak with your colleagues.

It makes a lot of sense. Most people actually don't live very far from the friends and the family members that they'd want to be in presence with. Do you think there's anything to that? Is it going to go enterprise first and then consumer? And if so, is it going to be like a weird position for Facebook to be like an enterprise technology provider, whereas like the social network might come sometime in the future? Well, I mean, I think you've nailed one thing here, which is, which is the enterprise use case is definitely product market fit like right now.

We have people doing meetings. I do meetings in NVR right now. And there are attributes of it that are way better than a video conference. And we have a lot of technology coming down the pipeline to make that even better, less friction, you know, more value. So like, I think that like meetings are going to be there. I think what we saw with Zoom is like virtual meetings aren't just for work anymore. So like I don't think there's this massive distinction between consumer and enterprise anymore. And so I think if it's like meeting with other humans remotely is like better in VR, I think you're going to

see consumer applications of that today. I mean, if you look at how people are using VR, social applications are a huge part of it already. Like so I think we're already seeing that sort of desire to new technology for connection, which we've seen throughout humanity. Yeah, well, with the Zoom, you know, not just for Enterprise. It was if you were in a lockdown, you might want to Zoom with your friends. Thank goodness. Most of us are not in that situation anymore. I never want to look at a friend through a Zoom window again.

Yeah, well, I have family that lives on the other coast. Like, you know, I have FaceTime with them. Like, yes, I would rather meet them in person, but that's not practical. And so I'm not saying it's a replacement, but it is an option. Yeah. Okay. We're coming in for a landing here. Two more questions. If you could change anything over the course of your, let's see, 14 years at Meta, Facebook, what would it be? One thing. Ooh, can you give me the second question so I can think on that one? Because that's a hard one. Second question is, what do we have to look forward to? Technology-wise.

At Meadow or in the world? Just put on your technology hat. What technology do we have to look forward to that we're most excited? I mean, I think, so I'm going to answer that one while I think about the other one is I think we have a ton to look forward to. I think we're going to have, you know, new ways to travel. So, not just electric vehicles, but, you know, boats that are flying over the water on hydrofoils with electric power trains that are smooth and fast and carbon-free. I think we'll have electric vehicles that will fly you from destination. So a bit of transport around. I think we will see robotics actually enter, you know, more and more applications over the coming decades.

I think we'll see AI move into realms that would be more obvious to the average consumer in the coming decades. I think this idea of ARVR as a commonplace way for people to communicate will be very clear in the next decade or so, especially as you look at advances in technology and avatars and others. So I think we have a lot to look forward to. And all of those things are, what I care about is like what makes people's lives better. And I think all of those, all of those things do. And I could go on for a long time. But those are the ones that I'm, I'm particularly excited about because they're just

like better with no compromise. And that's what technology can do uniquely, I think. In terms of, you know, what to do different. I mean, you know, I'm, and I know this may not resonate with people. It's like we, we poured our halt and soul. And I took our, you know, the world's best AI team and concentrated them on this, this content management problem from the very early days. And this is why most of our work you see in those results and not in other splashy demos. You know, and I, you know, I guess I wish we could

have done that even earlier because there's just a ton of work to do. And, you know, every bad experience people have on the product I, I feel terrible about and personally responsible for. So I think, you know, the more we could do to stop those bad experiences so people can just connect and communicate and have fun, which is what they're supposed to do, the better I feel. It's really always great to talk. I appreciate you spending time with me again and sharing so much of your perspective and being candid about the good and the bad. So thank you for being here. It's always fun. Thanks for taking the time. Yeah, we'll do it again soon, hopefully. Okay. Well, thanks again for joining us. Thank you, Nick Guatney, for doing the editing and mastering the sounds. Here we are. We're back on the horse again after the Davos chats. Thank you, LinkedIn, for having me as part of your podcast network. And thanks to all of you listeners for being here once again, joining us on a Wednesday or whenever you listen. on the big technology podcast. We will be back next week with another show with a tech insider or an outside agitator. Until then, take care.