125 points by anuh 8 months ago
1.) Does the machine learning algorithm behind toutiao abide by the censorship effort from Beijing? For example, are any of the contents about the disappearance of the wife of the nobel prize winner Liu Xiaobao censored?
2.) What happens when the Chinese government demand board seats, much like Tencent and Alibaba? Or is toutiao already working with the Chinese government closely?
3.) Do you know whether or not toutiao is allowing this technology to be outsourced to other dictatorship countries that would allow their leaders to effectively control the contents people read every day?
4.) Are you aware that toutiao is spreading its influence into East Asia, and possibly India, and is it possible that it would spread Chinese government's message through its apps?
> Or is toutiao already working with the Chinese government closely?
Every Chinese social media app/company has to follow the Chinese censorship rules, so yes. Toutiao has an army of censors working to censor the content.
Nobel Peace Prize is more political and ideological than any other Nobel Prizes. I don't consider it to be a 100% valid one. Not like academic ones where people send proofs.
And that guy is named Liu XiaoBo not XiaoBao. His wife is fine, and there are no reasons to do anything about his family.
Toutiao is just another company who runs for profits. Profits are focused. I don't see its popularity outside China even in East Asia.
You may worry more about your own right-wing Indian government.
"wife is fine"
Hu told BBC that three years of house arrest had thrown Liu into deep depression and a health professional had been prescribing anti-depressants medication for her. 
First of all, I wouldn't consider Toutiao content to be "quality content", as the article says. It's still pretty much all "soft news".
Secondly, this 100% engagement-driven algorism of content recommendation doesn't serve to inform readers in any meaningful way, as a proper News app should do. But rather, it just makes ppl keep clicking and wasting their time.
There was a whole chain of separate departments dealing with proletarian literature, music, drama, and entertainment generally. Here were produced rubbishy newspapers containing almost nothing except sport, crime and astrology, sensational five-cent novelettes, films oozing with sex, and sentimental songs which were composed entirely by mechanical means on a special kind of kaleidoscope known as a versificator.
-- 1984, George Orwell
This is what stuck out most for me:
> The engine learns quickly – for most users, it takes less than one day to successfully learn their interests (indicated by 80% read rates).
That is insane. I don't know of any other platform (except for maybe image heavy subreddits?) that have that rate of content consumption. Facebook obviously tries, but they are so focused on their limited content sources that they frequently don't have the content people want. Twitter has similar issues, if there aren't 50 tweets you want to read in a row, as you keep scrolling Twitter has to drop the quality of content down to make sure the feed shows something.
But by being willing to surface any type of content (give users what they want), it sounds like Toutiao has been able to fully leverage AI, removing constraints on what content exists naturally leads to better results.
Also, it is a miracle if they truly solved ad targeting. In comparison, I am still getting ads for a belt I bought 3 weeks ago.
> That is insane.
No, it's not. It's just easier to learn to entertain the average people, than to entertain the smaller group of more tasteful people, or to learn and sell things.
I was also quite impressed when I learnt the figures, and am still impressed at the fact that it managed to capture the mode of people's interest, but my every attempt to use Toutiao has ended in pain. I certainly spent longer than one day.
As an avid user of Toutiao, I am super impressed by Toutiao. It is super easy to use and shows the content I am mostly interested in. I use it daily to get information. I am a little bit worried that I would not get broad views if Toutiao only shows articles I am interested in.
Fliboard's recommendation engine is really good, I would say it has already solved the content discovery and recommendation piece of the puzzle. The key is to keep us engaged early on and let them keep using the product.
The main differentiator is the Toutiao is able to leverage AI to write good grammar article ... in Chinese when there aren't enough content already. The China has SO MUCH data required for deep learning due to its population and omnipresent of mobile devices. Other countries just couldn't compete.
Good point. I believe Toutiao tries to show different content ocassionally to solve the exact problem you are describing. There may be new areas or other hot trending topics especially locally that you may need to be aware of and they use these attributes to push content. But this is something they continue to work on as well.
just a little?
TL;DR; Does engagement equal success?
I have an account on Youtube that I used for only a specific interest.
Youtube's suggestions are pretty bad. And they've declined in quality. And Youtube makes a strong effort to selling me mainstream content that I uniformly reject. And I can find better recommendations on suggested videos fitting my interest than I find on the Youtube landing for my account. And I'd be for Youtube to keep a list of search terms and such for further viewing.
My deduction from this is that Youtube is more interested in pushing content of other people's choice than giving me what I want. And that's pretty standard for how all media has worked since the early days of radio.
So it seems like it would be easy for a content company to get engagement by giving people what they want to view. And the challenge problem is often what people want to view doesn't garner much payment per views. And so having a bunch of engagement by itself doesn't give you success, rather what makes money is careful balance of desired content and pushed-content (IE crap).
Very much disagree. Youtube's suggestion algorithm are improved incredibly over the years. It is increasingly surfacing content that I find interesting. I spend much more time on Youtube nowadays, easy more time than Facebook. I regularly just go to Youtube.com landing page to see what it wants to show me and always shows great quality content. Youtube has done an amazing job with their algorithms, and its showing.
Check out this article: https://www.theverge.com/2017/8/30/16222850/youtube-google-b...
Engagement alone is not success but sustainable and improving engagement (I.e. strong and improving cohort retention) is a measure of success because it creates a high barrier of exit for you as a user to stop using the app. 74 mins daily is not easy to create and sustain especially when you don’t have a social graph (On Facebook you are spending time talking to friends, reading updates from friends and content) but on Toutiao - you are only consuming content.
How many "social features" does Toutiaco have in comparison to Facebook or Twitter? I thought that was the essential feature to get user engagement, but Toutiao blows both Facebook, Twitter, Instagram and Snapchat out of the water.
Toutiao stays true to Asian usability dogmas and opens all links in new tabs (target="_blank").
Good question. They did not have many social features at the start - especially in year 1 (no followers, no friends, no chat). Over time they added like, dislike, retweet, comments functionality.What made them really stand out in the first year was the catchy title, delicate product design and pretty much the first personalized news aggregator with recommendations in 2012. They also iterated heavily to remain as one of the top apps in the app store during the first year. In the first month - users gravitated towards the app as it became the destination to read all top trending news /stories that were going viral that day (it solved a gap in the market, no one else was doing it). Over time they recommended stories to users and tracked data to ensure only relevant stories were being pushed which helped increase engagement.
Toutiao is solving the Internet's current most pressing consumer-facing problem - the discovery of relevant content.
It's a problem orthogonal to what Google solved earlier in the millenium, which is roughly "How do I find an answer on the Internet to a question I already have in mind?"
As the Internet has developed, the cost of content creation and data have decreased dramatically. And so the sheer quantity of content has exploded. Every consumer tech company has partially recognized this problem - Facebook, Twitter, Quora, Instagram, Snapchat, and Medium all want to recommend you relevant content, all have adopted feeds of articles and posts and pictures, and all desperately want your continued attention.
At this point, there's likely good content on the Internet for every single person's taste (like Anu said, there's a very long tail of content creators). The service that provides the most relevant selection of it will win consumer interest and usage at the largest scale.
And it certainly seems that Toutiao has leapfrogged everyone else in this regard. I'm very excited (but also a little wary) to see what they come up with in the future.
google absolutely sucks at it, probably because they don't see any value in it given their user base (everyone) and lack of competition.
i.e. my youtube front page just shows the same 100 video links over and over and over.
there isn't even a really good way to discover relevant content in the channels i've already subscribed to. it's pretty lame.
YouTube makes me want to punch my computer.
The related videos are never relevant, letting a child meander through YouTube invariably winds up showing my kids super inappropriate content. The YouTube mobile app sucks horribly (ios) I literally think that YouTube is one of the worst ubiquitous services there is.
This isn't a great solution, because kids are pretty good at getting out and into whatever app they want, but there is a YouTube Kids app. It's more like streaming TV than YouTube.
Toutiao is one of the names often bandied as a potential candidate for having a US-based IPO, alongside Best, Roku, CarGurus, and some 100+ other Chinese consumer internet unicorns ;)
Your insightful analysis of Toutiao's AI depth is enough to tempt one to dip a toe should shares trade publicly. But the revenue growth (contextual ad based) is phenomenal as well. Any clues as to how they might seek to expand into paid content or media? Or otherwise grow beyond Chinese language centric markets?
Chinese startup Toutiao raising funds at over $20 billion valuation
Toutiao is now focused on video as it is increasingly becoming more important (more supply of video content and more consumption as well). Outside of Chinese language centric markets they have an investment in India (which they have announced publicly). India is definitely a market that could potentially benefit from algorithm based content discovery (several different states/languages and a long tail of content that is not easily available to users).
Could you please link to the website of China's Cargurus that you mentioned? I would like to see which site you were speaking about because I follow Autohome and Bitauto.
Huh, More time-killing than social media. Feeding people content that they like. Putting them in the bubbles and reap the profits. Much of the Internet has been on this track for years. Last year's election fully testified how much all of this kind of media can do.
Reminds me of this https://www.theawl.com/2015/06/a-complete-taxonomy-of-intern...
Hey! Anu here, a Partner at YC's Continuity Fund. Happy to chat about anything discussed in this post and would love any feedback.
I'm a little disappointed in this post given China's penchant to controlling news/media/content.
I see how this could easily be manipulated into an extremely powerful propaganda tool - especially hiding behind a guise of "algorithms" are doing all the work - and yet it is not mentioned at all in the post.
We have already seen how content systems like Facebook, Google, and Twitter can be manipulated via botnets, but when a central power manages this it can obviously get significantly worse.
Does this concern you as an investor who lives in America?
I am afraid I cannot do justice to this question by answering it in a few sentences in this forum. Toutiao is one of the first apps in China to enable creators to write content and find audience without having to go through other media outlets. So in a way they are transforming the way information is made available to people. If you would like to discuss more email me directly at firstname.lastname@example.org
I agree it's tough to do justice in a few sentences in a forum, which is why I assumed it would be done in the very in depth and otherwise interesting article you wrote :)
In my opinion it's the absence of this topic from the article that makes YC look tone deaf once again by glossing over a serious repercussion that technology can have on society for the sake of self-promotion -- especially given current events regarding the manipulation of social networks by foreign entities and a very serious history of censorship and media control.
All media are controlled by the government in China, not just Toutian. It certainly can be manipulated easily for propaganda. However, it is now a common problem even for a company like the Facebook's fake news problem.
You talk here about creating an "addictive" product. People in the industry usually use the euphemism "engagement" rather than "addiction" but either way, do you think that this type of product will eventually be treated by society the way cigarettes are now?
Given that most people in SV don't have connections to China, how did your personal investment in Toutiao come about?
And given the amount of money in China, why did Toutiao need or want non-Chinese investors, particularly individual investors?
As someone not familiar with Toutiao, could you describe how the Q&A function works from a user perspective?
I'm guessing there's a part of the app where the user goes for Q&As. If a user asks "How do I flush my car's cooling system?", does the question get sent to all users who like cars?
Once a question is submitted, the AI engine automatically recommends a list of relevant experts/potential answerers to the questioner to choose from. As a user you can choose who you want answers from. In addition the system also pushes the question to relevant users based on their interest graph.
The system then also sorts the recommended answers based on its quality.
Has Toutiao been through YC core program?
Nope. I have known the company from before my time at YC.
>> The average user spends more than 74 minutes each day in Toutiao
so many wasted lives
I’d guess many of us spend 74 minutes a day on HN
Happiness is not waste
I'm sure there's research on this already, but just a thought - it seems to me like it would be easier to do NLP on Chinese, vs English. (I speak both). Much easier to recognize the author's meaning in Chinese characters, and more often than not people use the same words to describe something.. vs. the frankly disgusting mess that is English. Misspellings abound, the same words can mean all kinds of things, grepping for strings is hard because sequences of letters aren't unique, etc.
Do you speak Chinese? NLP for Chinese comes with it's own set of challenges.
Because word boundaries aren't marked, you need word segmentation, which requires understanding in lots of cases. (Can't tell where a word ends and the next one begins if you don't know what they mean.)
Many words have a literal meaning and a metaphorical one (e.g. 纠结 can mean both "tangled" and "confused").
Different synonyms being used for the same concept are common as well, especially when you contrast formal vs. informal writing.
Misspellings can happen too, where a character is substituted with a similar character that has the same pronunciation. Sometimes completely different characters are used as a kind of pun.
Grepping is less likely to yield false positives (unless you're looking for a single-character word that can appear in compounds), but there is no easy way to do fuzzy matching to account for misspelled words.
Most probably this might not get a reply but I still wanted to know about this statement -
"During the 2016 Olympics, a Toutiao bot wrote original news coverage, publishing stories on major events more quickly than traditional media outlets. The bot-written articles enjoyed read rates (# of reads and # of impressions) in line with those produced at a slower speed and higher cost by human writers on average."
1. What does writing original news coverage entail? Did the bot put together couple of news source to create an article or do you mean it transcribed the news from the live video feed as it happened?
2. How much of the bot feed works due to high declension? . Because normally bot structure and output tend to make less sense as the sentence structure comes out all wrong.
Additionally, there is quite a lot of focus on getting user to stay on the platform. Not to sound alarmist here but doesn't it mean you are most of the time helping people reinforce their world views? News should be reported as is but then if user engagement is so important news will be written right or left wing just to keep the user on platform.
First, writing stories on Olympic game results required data, and Toutiao pulled it from three sources: [a] real time score updates from the Olympics organization, [b] images from an image-gathering-company it had recently acquired to find relevant visual media, and [c] monitoring live text commentary about the game.
Second, Toutiao had to figure out how to combine data from these three sources to ensure an internally consistent and relevant story.
As a result of these efforts they were able to publish news story of a game approximately 2 seconds after the event ended as they collected and processed the information as the game progressed.
In Toutiao's case news is the element with which they launched but today there is a lot more reason why users use the app (much beyond news). Local news, weather, information on agriculture, etc. are surfaced regularly which is why the app is quite popular nationally in China and not just in Beijing and Shanghai.
AI just got a resurgence in the last couple years and it's interesting to see how quickly they have adopted the technologies to build a company. Simple AI techniques applied at key areas can generate huge result.
"While timing is everything for a startup, it takes deliberate effort to build an addictive app"
Anyone else deeply uncomfortable with this? Not a useful app, an addictive app. :/
The (Beijing?) smog pictured here is somehow more poignant since the article has 0 to do with pollution... just people living their lives. Scary.
That is Mountain Taijun, 7000+ feet peak with national AAAAA-rated tourist attractions (and not in Beijing). So no, I don't think that's smog, probably just fog.
Peak's Wikipedia page: https://zh.wikipedia.org/wiki/%E8%80%81%E5%90%9B%E5%B1%B1_(%...
Well that is certainly less dramatic than before. Thanks!
Not to discount their achievements, but it's important to note that as the content creator (instead of an aggregator or a content platform), it's much easier to have control over the quality of metadata, which is immensely useful in recommendation engines.
Checked out Toutiao's website, I think registering if you don't speak Chinese is nearly impossible - google translate doesn't seem to work on the login page.
You can copy/paste each string into google translate individually, if you're patient enough ;)
It is a mobile first app :) I don’t think they want anyone going to their website!
Hint: it's not aimed towards non-Chinese speakers...yet
> read rates (# of reads divided by # of impressions)
Why does Toutiao use this metric to measure quality of an article?
Not a measure of quality, but a measure of potential profit.
How can we trust any of the analytics coming out of mainland China? Can you be 100% positive that Toutiao is not controlled by the Chinese government? They can easily funnel all internet traffic to Toutiao and control it, and use it to their advantage. Curate the news to whatever way they want.
Toutiao is a privately owned company held in Beijing. Am I being paranoid? They can claim machine learning (and they probably do). However, until we can be 100% certain the Chinese government is not involved, I wouldn't personally trust any of the stats.
It's public knowledge in China that all tech giants are in bed with the government. Considering Toutiao is one of the fastest growing media companies in China, you can be sure they too will bend to the will of the government at a moment's notice.
> They can easily funnel all internet traffic to Toutiao
They can't force users to download the app and use it (without us knowing it), can they? They could control the content it shows users, but how does that make the stats inaccurate?
I can’t convince you to be a believer :) That is totally up to you!
Can you go home at the end of the day, look into your kid's eyes, and say 'mommy made a difference today, helping a dictatorship government control its own people, and spread its controlling message, eventually to the world'?
Perhaps to your surprise, many people in China are okay with certain freedom tradeoffs vs. improvements in other areas of quality of life. Its not an issue that you can easily simplify the more you familiarize yourself with it.
I hear this argument a lot, but:
1: If so many people are ok with it, then why not take a vote? If they are ok with it, they will vote for the current government. If they are not ok, then this argument is false and the current government should be voted out of office. Either way, a vote is good for China.
2: If the citizens are ok with it, then why does the government have to censor criticism and non-government ideas, run massive propaganda operations on their own citizens, jail dissenters, and oppress Tibet and Xinjiang? How do you know if the citizens are ok with it?
3: Parts of China that do have democracy and human rights, Taiwan and Hong Kong, seem to like those things very much - it's not a popular idea in those regions, or even an idea at all, to trade in democracy and human rights for the mainland's political system; no Taiwanese candidate makes that pitch to the voters! Also, people in Taiwan and Hong Kong are far more prosperous per capita than those on the mainland (I'm excluding Hong Kong from the "mainland" in this case.)
> certain freedom tradeoffs vs. improvements in other areas of quality of life.
It's not a tradeoff, it's just a needless sacrifice. All the wealthiest, most secure nations it the world have democracy and human rights. It's not a Western thing; that includes the wealthiest regions of China, Japan, South Korea, etc. What is the Communist dictatorship providing in this tradeoff?
> Its not an issue that you can easily simplify
'It's complicated' is not an answer, it's the avoidance of one.
Trade, not democracy, brings prosperity.
Hmmm ... trade is important, but it's hard to miss the overhwhelming correlation between democracy and prosperity.
Name a '1st world' nation that isn't democratic. But plenty of poorer nations trade.
One theory is that democracy provides greater political stability and the rule of law, and those are essential to prosperity. People are not willing to risk their life savings on a business when the government or some other powerful figure can just take it away, or if political instability will undermine their economic opportunities.
Plenty of poorer nations have some kind of democracy as well, although in many cases it's debatable how free their elections really are.
I think the correlation between democracy and prosperity is indicative of causation in the opposite direction: citizens of prosperous countries want a say in matters that affect them, and they have the time on their hands to campaign for their causes. Many countries have only become democratic when their economies were already prospering, e.g. South Korea, which was a military dictatorship for much of its initial growth.
I also don't think that democracy is the best way to provide stability, since the election cycle tends to create uncertainty on relatively short timescales. How are you going to build a business if the electorate might decide in a few years that they want to make it illegal, or impose stricter regulations, or demand higher taxes? For maximum stability, an autocracy with a slow but predictable bureaucracy would be much better. Of course, if that means your slow but predictable death, it becomes the much worse option, but few people seem to be concerned about that in China.
> few people seem to be concerned about that in China
See my original post, above. The Chinese government clearly is concerned that many people in China are concerned about it.
> I also don't think that democracy is the best way to provide stability, since the election cycle tends to create uncertainty on relatively short timescales. How are you going to build a business if the electorate might decide in a few years that they want to make it illegal, or impose stricter regulations, or demand higher taxes?
That's not how democracies work; they don't elect kings and queens, who can arbitrarily change laws. They elect officials to a wide variety of roles, all responsible to the electorate, who must agree on changes through legal processes. In the U.S., a majority of the House, of the Senate (and sometimes 60% of the Senate), and the President must all agree on a new law. Also, the law must pass muster in the courts, and be legal under the Constitution. Also, voters get a powerful say in what the laws are. If you don't like a proposed or actual law, you can speak up, build support, and influence the outcome - it happens all the time. Finally, the federal government's power is limited to federal issues; states and localities override federal policies or create their own policies all the time. You can see the current U.S. president's lack of success in his legislative agenda as an example of how democracies work.
In an autocracy, one person can arbitrarily change whatever they want - they are not subject to other laws, a constitution, representative legislative bodies, electorates, or any of those other influences.
And the outcomes are clear: All the most stable countries are democracies, and even when they were much poorer than today (such as the U.S. in the 18th century).
The Chinese government is concerned that more people will become concerned about human rights issues, not that there currently are many who care. Otherwise, censoring undesired opinions on the internet wouldn't be enough.
In your explanation of the way the US political system provides stability, the important factor is not that bureaucrats are elected, but that their process is slow. It would be equally stable if they were appointed by birthright or any other undemocratic process.
That voters get input into the legislative process is of course a good thing for them in aggregate, but the progress it creates diminishes stability. You can't just keep running your business the same way when the legal environment keeps changing.
Worse, the number of people who get input means that the outcome will be very uncertain. Maybe the law fails to gather support, or it is passed but later ruled unconstitutional, or it is amended after the next election cycle. You have to account for all of these possibilities. Things would be much more stable if you knew far in advance what will happen.
You have convinced me that a single person being able to influence everything might not be very stable either, depending on how fast that individual changes their mind. I now think that a large body of decision makers, appointed for life by any process (be it democratic, autocratic or random) requiring a large quorum of agreement for any decision, would be the most stable system.
If you really want to argue that democracy causes stability, your last sentence should have been: "All democracies are stable countries ...". But I'm not sure you could really argue that, considering the number of democracies who went through very unstable periods in their history. (E.g. civil war in the US.)
Requiring large numbers of people to agree to a change makes it predictable - I've seen few surprises from Western governments; whatever one thinks of Trump, his behavior is what he advertised in the campaign.
But whatever your theories, the data is that democracies do have the most stable and predictable business and legal environments, and more so than China where instability and lack of rule of law is a complaint of businesses, especially foreign ones. Almost all the world's leading businesses are based in democracies; the vast majority of business is done in democracies. Those monarchies that ran the world for millennia didn't perform so well.
It's not just predictability either; corruption is a huge problem. In places with lifetime appointments, what happens to corrupt leaders? They can't be voted out, and lacking the rule of law and separation of powers, they will not be investigated and tried (unless a more powerful leader causes them to be). And why should we think that lifetime appointees will care at all about the voters? The lesson is, power corrupts, and absolute power corrupts absolutely.
Countries without democracies, in fact, stagnate (such as China until the encounter with the West in the 19th century and again under the Communists (compare w/ Taiwan and Hong Kong), the USSR, N. Korea (compare with S. Korea) and many, many more). Also, by not giving citizens a forceful voice, they ignore and overlook citizens' interests, leaving serious issues unaddressed. Even the well-intentioned cannot centrally plan politics any better than they can centrally plan economics - nobody knows enough to understand and address the needs of hundreds of millions.
Finally, one reason democracies have more prosperity is that they are more dynamic. In autocracies, economic success tends to depend much more on access to those in power. The government will never allow a SpaceX to challenge the favorites of the powerful, Boeing and NASA; or allow Microsoft and Compaq to take down IBM; the Internet to take down the telecomm system (at least not until it was proven to be a necessity in the West), etc. It's because the autocrats have no reason to allow a challenge to their own power and every reason to prevent it.
I'm afraid this debate was settled long ago, around 1776 in the U.S., and monarchy and autocracy clearly lost.
I agree that overly stable countries tend to stagnate when that stifles innovation; I certainly didn't intend to claim that stability was universally good. After all, progress is impossible without the instability it brings.
However, I find it interesting that you mention the examples of PRC vs. ROC and North vs. South Korea as autocracy vs. democracy. In fact, after World War II, all four of them were military dictatorships/one-party states, none of them particularly inclined towards democracy. The only difference was their alliance to either the USSR (PRC and NK) or the US (ROC and SK). This helped ROC and SK to develop their economies (still under autocratic rule) and likely contributed to their eventual transition to democracy (while keeping their economic growth).
That would seem to indicate that economic development leads to democracy, and not the other way around. I tried to make this point in my previous comments, but you did not address it. I'd like to know what you think about that.
> I agree that overly stable countries tend to stagnate when that stifles innovation
That's not what I said. I said autocracies stagnate; wealthy democracies are the most dynamic, innovative economies, in business, technology, science, culture and everything else. That's where the Internet, web, Facebook, the startup business model, financial engineering, bitcoin, globalization, open source, rock'n'roll, hip hop, electronic dance music, personal computers and smartphones, modern 'deep learning' AI, and almost any other innovation you can name comes from. It's where science, capitalism, democracy, universal human rights, and much more came from.
Couldn't government-controlled company publish their real stats?
I personally think any sufficiently large company in China is partially government-controlled. But does it change any of the business analysis presented here?
What would be the closest app equivalent in the western world to Toutiao?