Like many of you I have a bunch of Google alerts set up on certain topics, with one of them being TikTok. I’m not even on really on TikTok, but I began tracking it a while back because all of the hype surrounding it, and then when a friend of mine told me about how his son had only been on it a couple months – doing a few videos making different things out of boxes – had over 60K followers – I got even more curious, which is why I did a conversation with him for this series not that long ago. By the way he’s now over 70K followers….
But then last month I got a series of Google alerts about something I hadn’t seen coming – that TikTok had surpassed YouTube in viewer time per user. And that really grabbed my attention, as I watch a bunch of YouTube videos for a variety of reasons. And with my interest growing in understanding the draw TikTok was having, my CRM Playaz co-host introduced me to, without a doubt, the most thorough and insightful series of posts on TikTok that is out on the web. And after reading them we both wanted to speak to their author, consumer and entertainment technology expert Eugene Wei, to dig in a bit more on what TikTok is doing and how they’re different than the other major social networking platforms. So below is an edited transcript from a portion of the conversation Paul and I had with Eugene.
I know this is longer than the usual conversation transcripts, but there is so much good stuff in here about TikTok in other related technology areas that I didn’t want to cut anything out. But this is still just a portion of the hour plus conversation. So if you are at all interested in how TikTok works and why it’s driving the numbers it is putting up, check out the full convo by clicking on the embedded SoundCloud player…and learn something. A lot of something.
The Role of TikTok as a Communications Medium
Eugene Wei: I think we probably are living through a phase change, in terms of the relative power of different media. Not to say that video, hasn’t been a super important medium in the past few decades in America, but it was largely through a broadcast configuration with a few central gatekeepers, TV networks, movie studios, dominating the use of that medium. And what the smartphone and social media and smartphones with video cameras did is what the internet does to everything in the world. It changes it from a gatekeeper hub and spoke type of model to a network configuration where every node can also broadcast out. And the big thing, there’s this quote by the naturalist, E.O. Wilson, which I really like, which I think I mentioned to you last time I spoke.
Paul Greenberg: Yeah, you did.
Eugene Wei: It goes something like, “The problem with modernity is that we have God-like technology, medieval institutions, and paleolithic emotions.” I think that’s how it goes. It’s a great quote.
Brent Leary: That is.
Eugene Wei: But the thing I say, which is my very specific variant of his quote, is that the strange thing about the modern world is that we have God-like search and distribution and paleolithic emotions. So what social media did for the average person was for the first time in history, any person could put anything out in the world and potentially have it reach a billion plus people in the world, which that’s pretty much one of the most momentous shifts in communications, in the history of the world. Obviously, with the printing press, some of the number of people could suddenly reach a lot of other people, but it was still pretty gated. And then even in the past few decades, prior to the internet, yes, one person could reach a lot of people, but you have to be accepted by a publishing house, a music label, a movie studio. They still kind of controlled that.
Brent Leary: Pay a lot of money.
Eugene Wei: Yeah, that was one way. But now anybody can. You just have to learn how social media algorithms work and they’ve connected up so much of the world and have these algorithms that choose what gets pushed out. And TikTok, in many ways, took that to the umpteenth degree. Prior to that, you had things like Instagram and Facebook, where if you wrote something that was popular, it could move fast, but it was still gated a little bit by the fact that those social networks are built around follow-graphs. And so if you didn’t have a lot of followers, you had to rely on people with more followings to share your thing out. There was still some limit to how many people could see it. TikTok came along and said, “If you make a great video, we will just show it to people. They don’t have to follow you. We’ll just show it to anybody we think will find it interesting.”
What does God-like distribution look like?
Eugene Wei: Theoretically, if you made something generically popular, they might show it to everybody, which is a staggering amount of people. So they took distribution and they said, “What does God-like distribution really look like? Well, let’s put this in the heads of a hundred million people in a short period of time.” When Charli D’Amelio, who was one of the early TikTok stars, started gaining a following, everybody was… You could see it in your For You page, the main feed of TikTok, people are like, “Wait, why is this girl becoming so popular?” And people were arguing over it. And then that just made her more popular and pretty soon she had a hundred million followers on TikTok.
I don’t know how long it took her to go to a hundred million followers, but she might be one of the fastest social media people to get to a hundred million followers in history, if you compare it to Facebook and Instagram and Twitter. It seems like the cycle just accelerates more and more. And so what we’re living in, in the video media world, is really this networked, turbocharged world where information gets really put through a rail gun and just accelerated.
In a way, that’s different from the broadcast era of history. And so I think, if you’re looking at gen Z, you’re looking at a generation of kids who’ve grown up in a more video dominant network configuration and will just take that for granted. But even look at the impact on old media. You guys know Bill Simmons the popular-
Paul Greenberg: Yeah.
Brent Leary: The Ringer, yeah.
Eugene Wei: I followed his career from the beginning. I was reading his columns when he had the AOL newsletter and then he moved to ESPN and then he eventually got his own website. And now what does he do? He’s doing podcasts, almost exclusively. He barely writes anymore. So you see, even someone who grew up in the previous internet era, they are shifting strategies completely.
Even all of us now talking, you said you used to write more and now you are moving more to video.
Paul Greenberg: Yep.
The Video Era as an Influencer Channel
Eugene Wei: I think everybody is seeing, “Oh, okay.” There are certain advantages to video over text. And I love writing. I love text as a medium. It’s certainly probably made me more popular on the internet than I would have been otherwise. But you have to admit, and I went to film school about this, so the video and text are different mediums and they have different strengths and weaknesses. One of the strengths of video, and even podcasts and audio, is that once you hit play, it just keeps going. The problem with text is in order for texts to work, someone has to read it.
They have to actually keep moving their eyes and processing. And so if they stop, the thing stops, but you can listen to a podcast while you’re doing your dishes or whatever. It’s just going to keep going. Same way that TV, you stop. People used to be astonished that the median viewer of TV would watch, I don’t know, five hours a day or whatever the stat was. It’s still some staggering amount. And people were like, “Who watches TV for that long?” But a lot of the time TV was just on in the background-
Paul Greenberg: Right.
Eugene Wei: As just like this passive medium, almost like there’s some degree to which you go see an exciting action movie, it’s very stimulating. You’d go see a horror movie, it’s very intense. But there’s a way in which the medium, if you just leave it on forever, it becomes almost like a narcotic. It just starts to just seep in your unconscious and-
Brent Leary: A hypnotic and a narcotic.
Eugene Wei: The opposite of stimulating in some ways. So, yeah, I don’t know. I feel like it’s… I know as a kid, I think if I had grown up in the YouTube era, I definitely would have been watching a lot of these vloggers every night, but I didn’t have that, so I watched TV or I read books, other mediums. So we’re just… I think these things, they battle out in the marketplace. So all these mediums are fighting in the marketplace. They’re all trying to get everybody’s attention. And the ones that float to the top will be the ones that we all choose as an audience. If we gravitate to certain medium, advertising dollars will flow to those mediums and those will prosper and you’ll have more people then go onto the supply side in those mediums.
TikTok, in a way, if you just look at that as a specific example, a lot of their business model and future will depend on them first winning in the marketplace of attention because we know attention is still finite. And then if they do so, they have a ton of optionality, because as you said, you can use short video to do any number of things. You already see brands on there trying to make hip little TikToks to advertise their products. We know that’s already working well in other markets in the world. There’s no reason it wouldn’t work in the largest ad market in the world, which is the U.S. They have tried, in China and other places, to work on education as a market.
Paul Greenberg: Yep.
Eugene Wei: So they’re going to pursue a whole bunch of opportunities that all come from the fact that you get those opportunities, if people are watching your app so frequently during the day. I recently saw this survey of incoming class of Harvard students. It’s one survey of one group of kids, so take it with a grain of salt, but something like, I think it was 10% of the incoming class or 20%, something like they watched two hours of TikTok a day or something like that.
Paul Greenberg: Wow.
Brent Leary: Jesus …
Eugene Wei: It was a staggering amount. And when I first… Before ByteDance had even bought Musical.ly and turned it into TikTok, I was in China and I met with a bunch of people who were telling me about Douyin, which was their knockoff of Musical.ly at the time. It was the TikTok of China. And people in China routinely would tell me that they had to uninstall Douyin from their phone because they were losing so much time, hours of productive time each day. And they were like, “I couldn’t even get any work done because I was just watching.” And Douyin in China’s even more dominant as a medium than TikTok is in the U.S.
Paul Greenberg: Oh God.
Social Media – East vs. West
Eugene Wei: The U.S. is such a developed and advanced media market, it’s just super competitive. I can even remember in my memory going back to China, and there only be really the three state-owned TV channels broadcasting, boring programming. And so for something like TikTok to come into that market, you’re like, “Oh, wow, this is completely… This is way more entertaining than anything we had on national media.” So, yeah, I think, in many ways, you can just look to China and say, “That’s a market that leapfrogged a little ahead to short video as just a medium for many purposes.” If you look at the Yelp of China or you look at dating apps in China, you look at any type of app that has a counterpart in the west, the Chinese version has way more short video in it. Largely, their internet grew up in an era where all their smartphones had video cameras on them. So it’s easier for them to just leap ahead and say, “Oh, okay, we should be leveraging short video more.”
Paul Greenberg: The valuation side is not actually as stupidly funny as it sounded at the time when I was reading it. What do you think of that?
Eugene Wei: I’ve long said that an interesting thing about the West and our social media is that almost all of them chose to build entirely around their unique social graphs. Twitter, Instagram, Facebook, they’re all built around this feed. There like a feed of content that you look at, and how do they source stories for that feed? They look at who you follow and they look at all the stories from the people that you follow. And then they try to pick the best ones of those to show you. What it is, is a feed built on a social graph.
But we know that people have interests. The weird thing is that the West largely try to guess at people’s interest graphs based on a follow graph. You follow these people, so you must be interested in what they publish. But anybody knows that our interests and then the people we know don’t align perfectly. And so there’s a mismatch in some ways.
I think Facebook is the best example of this, because they have, in many ways, the largest social graph in the West, where over time, they had to start putting in algorithms, because it turns out not everything from people you know is interesting or will interest you.
Paul Greenberg: For sure.
Eugene Wei: If there’s a lot of stuff that doesn’t interest you in the feed, you’ll stop checking the feed, and then they won’t get eyeballs on their ads, and so their entire business model is about trying to keep the feed as relevant and interesting for you as possible. And so the algorithm, in fact, is an error correction device. It is imposed on top of your social graph to try to remove the noise and improve the signal in your feed.
But if you think about it, it’s just odd that we try to approximate the interest graph with the social graph. Why not just build an interest graph? In many ways, that’s what TikTok is doing. They let you follow people, but they’re like, “You may follow these people, but if they publish a bad video, we don’t want to show it to you. We still want the videos to be high quality.”
I follow Charli D’Amelio, because I’m interested in seeing how the superstars of social media use the medium and interact with kind of the algorithms to try to game them. But I haven’t seen a video from her in my For You page for, I don’t know, months now. I know she’s still publishing videos, but I just never see them.
And the same thing, if you go to profiles on TikTok all the time, you see a video and you’re like, “Wow, this video is pretty good. I wonder if this person has made other popular videos?” You can click through on their profile, and you often see that that person has one TikTok hit and a whole bunch of videos that have hardly any views.
This is because TikTok, when it gets a new video, puts it through kind of a screening panel. They get a video from someone and they run it through some visual processing to try to identify objects and things. They look at the tags that you put in the caption. But they still don’t know if the video is any good. They may know some of the objects in it and some of the subjects, but they want to know if it’s good. What they do, they test the video on a thousand viewers and see what happens, and if that thousand-person panel doesn’t really react to the video, or they just scroll past it, that video gets buried by the algorithm. They’re not going to show it to anyone in their For You page.
Paul Greenberg: Wow.
Eugene Wei: I worked at a film festival once screening entries, or if you’ve ever had to be a judge on any sort of contest, you know that 99% of the entries are of really low quality, and that’s your job as the panel, to screen out the best from that. It is staggering, the volume of just terrible short films or films that I’ve seen working on that. I’ve learned to never take that job again. But this is just the way of the internet, right? We have more content, most of it’s terrible, and how do you screen? How do you curate that stuff out?
TikTok, essentially, has a really good, call it a kidney or whatever, just filtering out all that stuff and just showing you the stuff that has generated positive reactions from some group of people.
Paul Greenberg: Wow.
Eugene Wei: But TikTok came along and said, “Well, we have not given up on recommender systems for short video, though we recognize there are challenges.” Just the same way there are challenges in recommending podcasts. Whereas music recommendations have largely been cracked, right? If you just pick a song on Spotify and say, “Just based on this single song, generate a playlist.” Those playlists are pretty good. I frequently just use that.
But they’re still trying to crack the code on podcasts. And you say, “What’s the difference? They’re both audio formats.” Well, it turns out there are a lot of differences, qualitatively. Just the same way that recommending movies are hard. Movies and podcasts are really long. You tend to only watch or listen to them once. A podcast can be about multiple topics within the span of an hour or two-hour thing. A lot of whether a movie is appealing to you is about how it’s about, what it’s about. There are very subtle differences in why someone might like a movie or not, even if it’s within the genre that they like, so that’s a hard problem.
Follower-graphs vs. Interest-graphs
TikTok, with these weird videos, is like, “How do you crack the code on recommending the right videos to them?” Well, one, you have to know what these short videos are about. Two, you have to understand how this person feels about the video, but without making that process too burdensome on them. They did something which, whether it’s by design or whether it was by chance, is pretty brilliant. In seeing an algorithm, which was my second piece on TikTok, I wrote about how the entire design of the app is meant to watch you watching the video. Not watch you literally. They don’t turn on the camera. I don’t want people to get paranoid about that. They full screen the video, first of all. It’s on the screen at TikTok. Unlike other social media in the West, there’s only one thing on the screen. They do that because they need to know how you feel specifically about that one video. Whereas, Twitter, Facebook, Instagram, you have so many things on the screen at a time that it’s hard to tell what the person is reacting to.
TikTok now knows, okay, you’re looking at this video. Then they look at everything that you do. Do you let the video loop three times? Do you tap the like button? Do you share it out to somebody? Do you follow that person? Every positive signal is tracked. They can also track negative signals. If you just don’t even let the video finish and you scroll past it without reacting to it, that’s treated as a negative signal. And so now they’re getting very accurate feedback. When I say they show that video to a thousand people, like a test panel, I don’t know how many they show it to, but they show it to a small number of people. They’re looking at all these signals. And it turns out even by just testing a video on a thousand people, you get a pretty accurate read on whether the video is going to be a hit or not.
What we’ve learned about machine learning and machine learning recommendations algorithms in the past decade, I would say, one of the big lessons is with enough data you can achieve these phase shifts in quality. You can achieve these breakthroughs. It’s like text translation or text generation, text prediction. A lot of those in the early days were laughable. Everybody has funny stories about using Google Translate or Babel Fish back in the day to try to email someone in another country, and people just laughing at what it came up with for them. And then one day it’s good, and then the people are saying, Oh, machine learning can’t beat a game like Go or chess,” at first, and then it beat chess, and then it quickly conquered Go.
What happens is when these phase shifts happen, we’re all just astonished because it’s like bad, bad, bad, bad, and then great. It was the same with TikTok, right? The video recommendations, I would largely say people are very skeptical about the quality of video recommendations. While I don’t think TikTok shows the way to crack the code on movie recommendations, TikTok really did crack the code on short video recommendations. They generate a closed loop feedback system, which is very, very efficient.
This is part of the One-on-One Interview series with thought leaders. The transcript has been edited for publication. If it’s an audio or video interview, click on the embedded player above, or subscribe via iTunes or via Stitcher.