Scaling Up Censor-Friendly Content Aggregation
Author’s note: This is from the digital archives and I wrote it a few years ago. This was back when I thought that I could maintain a YouTube channel and a Medium blog at the same time. LOL, nice try Jon.
Looking back at it, I think there is very little that I would change. The main idea remains that content curation and “cleansing” is extremely expensive. I expect it is going to get more expensive going forward, and is a big reason why ByteDance has not yet filed to go public. When they eventually do, I reckon that “content censorship” cost is going to be quite large.
Something that I want to announce. I recently went onto a podcast as a guest. Check it out below. There are not many English-speaking podcasts in Taiwan, so I haven’t done a lot. But this one was
There is one thing that bothered me afterwards and good thing I looked it up. At one point I went on a tangent and said that the “guy at Intel who messed up the mobile transition” was a Ph.D.
I got to own up here. I done goofed. I was obviously referring to Paul Otellini, CEO at the time when Apple was building the iPhone. He was not a PhD. He had an MBA from Berkeley, my alma mater, but no PhD.
Sadness. I am aging and turning senile. Still, listen to the whole thing. It was super fun and for Chinese speakers who are learning English as a second language, you can do some learning too.
In January 2019, Bloomberg reported that ByteDance, the world’s most valuable startup, told investors that revenues came in at the lower end of its forecast.
Some have speculated that this is due to a slowing of the Chinese economy but I am curious about how much of the revenue slowdown is related to the company's newly “reinvigorated” need to adhere to Chinese content censorship standards.
It goes to show that being a content distributor — never easy in even the most progressive of countries — must be especially harrowing in China.
Introducing ByteDance
You might not have heard of this Chinese company but you may have heard of one of its apps including Toutiao or Tiktok or Musically. The company erupted onto the Chinese technology scene in 2012 with its breakout app - Toutiao. This app works like the Facebook News Feed in that it draws content from all across the internet and shows users the content its algorithms think they would like.
Some 120 million use the app today — most of them young millennials in large urban areas. Each user reads about 38 articles a day, spending 76 minutes per user per day — in 3Q 2018, Facebook reported that users spent an average of 79 minutes in all of its apps. ByteDance makes money from those users with ads kind of like with Facebook.
In November 2018, investors SoftBank, KKR, and General Atlantic led a funding round that valued ByteDance at $75 billion which makes it one of if not the largest startup in the world. This is higher than Uber’s last stated funding round of $72 billion (though a subsequent round valued it at $62 billion). It does seem to be less than the rumored $150 billion round that Alipay owner Ant Financial raised in June 2018 though.
Regardless, this is a really big company — $75 billion puts it ahead public companies like FedEx and Activision Blizzard.
ByteDance’s emergence has had ripple effects across the Chinese and American technology landscape. In China, the fact that Toutiao became a hit without the need to ally with either Tencent or Alibaba has disrupted their traditional rivalry.
It is especially vexing and alarming to Tencent, owner of the famous Chinese super-app WeChat. Time spent on Toutiao is time not spent on WeChat. Tech giants in America have noticed too. It has led to Google bringing Google News out of the company's backwaters into app format and Apple News to beef up its own News App.
ByteDance’s Strategy
ByteDance uses natural language processing and computer vision to discern what a video or web article is about as part of deciding as to whether to show it to a particular user. As Toutiao’s growth accelerates, the company has been leveraging those experiences in NLP and computer vision to bring artificial intelligence into other apps.
The most successful of those apps is Tiktok, which has been blowing up across America and has a huge marketing presence here in Taiwan (I see it pretty much in every YouTube video ad that I get).
You open the app and it immediately starts playing a short-form 15-second video. You can swipe to the right to see more videos from that creator or swipe down to see another AI-curated video. It is like what Vine could have been had it not sold to Twitter and withered.
There is an interesting article from venture fund Andreessen Horowitz that claims that TikTok is the flagship of a B2C trend in which artificial intelligence is the product itself.
There is no feed where you can scroll through and pick out a video that interests you like with YouTube or Facebook — you get just one video at a time. My guess is that they are tracking every swipe and every time metric as an indication of how engaged you are with the video.
ByteDance is basically betting that their content curation algorithm is so good that it can keep you watching and swiping. This is probably a good bet.
Tiktok’s growth seems to tear out a page from Facebook’s strategy book. Just like how Facebook used its social network to grow Instagram and Messenger, ByteDance will leverage their expertise in content and AI to essentially “take over” the phone screen with a constellation of apps.
Some of the other apps that ByteDance has released include Xigua Video, Hypstar, and Wukong Q&A.
The Fly in the Ointment
The trouble with this AI-driven content strategy is that you need a lot of content. At first Toutiao scraped the web for its content but over time it struck deals with publications to supply much of that content directly. And then after that it started to promote “influencers” to generate content for its users.
If you are familiar with the Facebook story of 2017 and 2018 then you can start to see where this is going. In time, shocking and objectionable content is going to flood into the system and ride those algorithms to prominence.
Some of this content is objectionable under any context. The New Yorker mentions some of what has been going around:
Daredevils scale skyscrapers (one climber was broadcast falling to his death), and young girls inject their faces with off-brand dermal fillers. In an infamous video series, a woman appeared to eat light bulbs, live goldfish, a cactus, and mealworms … Earlier this month, videos posted by girls reportedly aged fourteen and fifteen, in which they documented and glamorized their pregnancies—sometimes competing to be the youngest mother—went viral
This stuff is going to make some people uncomfortable even in America — and to think that this stuff is coming out in China of all places.
In America if this happens then you as CEO might be able to take a seat in front of Congress, use lawyerly semantics to dodge ham-handed queries from tech-illiterate septuagenarian politicians, and avoid lasting regulatory consequences. The Party focuses less on method than it does on outcome and as long as the outcome continues to displease the Party then the Party is going to do things to you.
So thus in April 2018, CEO and founder Zhang Yiming posted on his public Weibo and WeChat accounts an apology for failing to take his company on a path in line with socialist values:
"Our product was not in line with core socialist values…and we did a terrible job in guiding public opinion.”
Toutiao was temporarily yanked from app stores and the Chinese media regulator permanently removed one of its jokes apps Neihanapp. The founder also pledged to strengthen editor in chief responsibilities, boost the number of human staff reviewing content, create a blacklist, and more.
It is an extraordinary public letter. Someone can make an argument that China might suffer from the same public-private regulatory capture issues that plague western economies today. But this letter very clearly underpins the Party’s pleasure as the fundamental reason for ByteDance’s existence. Namely, it reads like a last warning.
Consequences
By far the biggest financial impact is going to the addition of 4,000 operational review staff for the review and removal of objectionable content.
Part of the reasons why ByteDance did not beat sales forecasts was due to the delay of monetization on certain functions and products — I am probably guessing that would be Tiktok. It also makes sense to think that a lot of this has to do with bulking up the processes and properly training 4,000 additional new staff. I would be very curious to see just how much of a profit hit this sort of effort is going to have on ByteDance’s bottom line.
Let us look to Facebook again. At the end of 2017, Facebook spent $5.6 billion on R&D and selling/general costs. The next quarter they added $900 million in additional costs, cutting the operating ratio from a stellar 56.5% to 46%. The quarter after that, another $800 million in costs on top of that (44% ratio).
And another some $600 million at the end of 3Q 2018 for a 41% ratio. In 9 months of 2018, Facebook has poured in an additional $2.3 billion in costs per quarter while only making $800 million more in revenue per quarter. $2.3 billion in additional expenditure still gets the CEO hauled in front of Congress.
ByteDance is reported to make $7 billion a year, so between that and the costs of marketing its apps I wonder how profitable the company is going to be. VCs are pouring money into this company because they see something they think is scalable — digitally harvested news paired with a sticky AI. But what if for every couple eyeballs you pick up through very scalable algorithms you got to spend on another (expensive) pair of eyeballs to censor content for them? That sounds not very scalable to me.
So with that in mind it is interesting to see this particular bit of news. In December 2018, ByteDance told reporters that they are looking to bring in half of their revenue from overseas.
The same content algorithms that worked so well in China can conceivably work outside of it too. There, you have less censor-happy regimes and weaker content regulations. Those revenues are going to be a lot more profitable, for now.
So, it might sound cynical but if I was running that company I would try to go overseas as fast as I can.