Feature Thursday: YouTube What To Watch

Can't do feature friday this week because Arnold suggested a great fun friday for tomorrow. So we will do feature thursday instead.

I got an update of the Android YouTube app this week and when I launched it, the home screen looked like this:


That is just the first two videos of the "what to watch" feed but you can scroll forever and the recommendations are awesome. After Matt Cohler and Mark Zuckerberg, I got the top 10 Knicks plays of the 2012-2013 season, an Arctic Monkeys video, and a video about a 3D printed prefab house.

This is like the #discover tab on Twitter but for video. It's awesome and I've added the YouTube app on my Nexus7 to my daily content consumption flow now. I don't watch every video, but I watch a bunch of them. I think this is a killer feature.

Strangely enough I don't get the same recommendations when I visit YouTube on the web. I get recommendations but they aren't as good. I wonder if they've rolled out something new on mobile first and will bring it to the web shortly.

I spent some time a few months ago setting up some subscriptions in YouTube and that is certainly a big reason for why the recommendations for me are so good. I would recommend doing that as well.

In any case, if you haven't checked out the new YouTube app and the "what to watch" feature, I strongly suggest giving it a whirl. 


Comments (Archived):

  1. andyswan

    I get nothing but barbie and mermaid recommendations. Daughter positioning herself well for own iPad as we head into birthday wish-list season.

    1. JimHirshfield

      You can’t fool us Andy. Your secret is out. And we support you.

      1. Matt A. Myers

        This will be Andy’s persona on a future AVC character t-shirt…

    2. andyidsinga

      hehe ..we had that problem big time on netflix. just switched to using he new profiles feature ..hopefully solves some of the ‘kid trash in the stream problem’

    3. fredwilson

      That’s great!

    4. Dave Pinsen

      I post form checks of me doing exercises to YouTube, to save space on my phone. Occasionally, among the recommended videos that come up, have been ones of women doing the same exercises with more weight. Thanks, YouTube.

    5. ShanaC


  2. takingpitches

    Perhaps the result of merging all the different sources of data across Google services#upsideofrunningthetable ๐Ÿ™‚ ?

  3. William Mougayar

    Cool. Isn’t that also what Shelby.tv is trying to do? I’m receiving recommendations from them via email. I did notice the YouTube app is way better on Android than on the iPhone because of how it lives on your home screen.Interesting that Twitter also recently launched Related Tweets, and after the Tech Circles discussion the other day, it seems like every Web service or App is getting into that.

  4. JimHirshfield

    It’s possible your logged in with different identities on mobile versus desktop. I know that often happens to me.

  5. JimHirshfield

    Arnold, fess up. The suspense is killing us.

    1. John Revay

      Inquiring minds want to know

    2. Donna Brewington White

      Setting my alarm early.

      1. JimHirshfield


  6. John Revay

    I’ve added the YouTube app on my Nexus7Fred – did you go out and buy a new Nexus 7 yet

    1. Richard

      great device but what’s up with the screen capture? You need three hands!

    2. fredwilson

      YesI love itTyping this on it now

      1. Girish Mehta

        Tangential…but interesting Nexus discussion a couple of days back by Benedict Evans. http://ben-evans.com/benedi

  7. falicon

    Nice! A friend recently got me hooked on the Playtube app. It lets you download youtube videos to watch offline – but the trick is knowing what videos you want to download for later…so this new recommendation thing is great and just what I wanted!

  8. reece

    like many platforms, you get what you giveif you do the work of subscribing to channels, then yes, YouTube can provide some solid recommendationsbut really, i think that the content on YouTube is just getting better and better. whether it’s YouTube creators, new media companies, or legacy content coming online, everyone accepts that online consumer video is legit and thus, we all benefit

    1. Greg Isenberg

      Agreed, Reece.

  9. Richard

    Out: CBS et al. Broadcasting In: Youtube et al. Narrowcasting

    1. Matt A. Myers

      Narrowcasting. I like that. Is that a new term?

      1. Richard

        Hit me about a decade ago after hearing a great keynote by Michael Saylor (founder of Microstrategy).

        1. Matt A. Myers

          Would be a catching term to use in a blog post title..

  10. Ray Gobberg

    @fred what are some of the channels that you rec for subscription? avid consumer of your blog and would love to hear some of your tops ๐Ÿ˜‰

    1. fredwilson

      NBA and Hoopmixtape

  11. reece

    also, we’re now alpha testing our own video recommendations @ Shelby.TV (cross-platform with three different proprietary graphs)if anyone wants to test, on web or iOS, hit me up

    1. ShanaC

      i do – i have a hard time finding shorter documentaries

  12. Dale Allyn

    As @reecepacheco:disqus mentions, one’s participation in subscriptions makes a big difference in recommendation quality. Are you throwing in the towel regarding your attempt to “de-Google” yourself a bit?

  13. Sebastian Wain

    Sorry for this inappropriate comment but I think the NSA recommendation engine will work much better!

  14. obscurelyfamous

    What to Watch in YouTube is eerily good. I never used to browse to YT directly, but I’ve opened up the app a few times and I’m immediately sucked into videos of stuff that I really want to see — and I don’t even follow many channels; that’s the scary part.Must be based the videos I’ve stumbled across in the past. I’m ok for spying on my Internet habits if it results in sweet videos.

    1. William Mougayar

      Yes! Take that analogy…and can you implement it on Disqus ?I want to open “My Disqus” and let it tell me what interesting Conversations are happening where it knows that I’ll be interested- both surprise me and keep me informed.

      1. sigmaalgebra

        Working on it.

  15. jason wright

    isn’t it a bit too much the self-reinforcing feedback loop? where’s the chance to break out and discover new things, those new things to read and discuss?

  16. sigmaalgebra

    Yes, discovery, recommendation, curation,notification, and subscription constitutea problem that needs a much bettersolution for maybe 1+ billion Internetusers and about 2/3rds of the content ofthe Internet, especially content not basedheavily on text.As in the threadTech CircleAug 20, 2013http://www.avc.com/a_vc/201…progress is needed even for content, e.g.,blogs, based heavily on text.Yes, there are various simple approachesthat have some utility, but it appearsthat a much better approach has been slowin coming.There are many ways to proceed that arenot very good.Mistake 1. Try to figure out what aperson likes and give it to them.Wrong. Are asking essentially a poorlyformulated question and can’t work well.Actually, can work for ad targeting, butreally can’t work well for what a personlikes. Why? E.g., I like classicalmusic. So, given a piece of music, do Ilike it? Too little information to say;maybe under some circumstances; likely notunder some other circumstances.Instead, just essentially ‘focus’. A leadinto how the ‘focus’ has to work is theword “interesting”. So, change that to’interest’. So, the user has an’interest’; now are on the way tosomething that can work well. At onetime, the user wants content for aspecific ‘interest’, but ad targeting canwork at nearly any time for any of theuser’s interests. An appropriatetheoretical and operational definition ofan ‘interest’ has to be a bit tricky toformulate.Mistake 2. Using the ‘social graph’ forfinding stuff is a total bummer if only asan especially weak special case of Mistake1. The accuracy will totally suck.Mistake 3. Using ‘leaders’ is also a poorspecial case of Mistake 1.Mistake 4. Using what’s popular is amistake. An extreme case is popular musicwhere what’s popular would returnessentially just some Top 40. That onesize fits all is a poor fit for nearlyeveryone. So, right, as in shoes, a keyis a lot in ‘personalization’ — easy tomention; so far tough to know how to do.Mistake 5. Getting content from similarcontent also has to work poorly. Firstneed a good measure of ‘similarity’ whichhas some severe challenges. Second, thegiven content is too little information topermit good results.There are more mistakes possible.We know, implicitly already, thatkeywords/phrases with a gross popularitymeasure such as ‘page rank’ work poorlyfor the purpose. A key reason is thatwhat is wanted is content with the desired’meaning’; keywords/phrases make hash outof meaning’; what’s popular takes us backto the mistake of the Top 40.Yes, writing software to work effectivelywith ‘meaning’ seems to be a holy grailproblem in computing. One approach is totry to write software that works withmeaning just like humans do — I’ll checkback in maybe 50 years. But the wayhumans work with meaning is not the onlygood way possible.Is there a good solution? Well, uh, …,maybe, …, uh. Uh, yes, might be somemath involved. I.e., how to takeavailable data and do manipulations toconvert it into good recommendations. Themanipulations will necessarily bemathematically ‘something’, understood ornot, powerful or not. Some powerfulmanipulations, well understood, are morepromising than shots in the dark. Notnecessarily elementary math, and maybesome math that’s original. The problem isquite general and short on mathematicalassumptions; so, some relatively general,advanced, and abstract math stands to beimportant.No, sorry, not ‘intuitive heuristics’,’artificial intelligence’, ‘machinelearning’, ‘collaborative filtering’,’support vector machines’, ‘singular valuedecomposition’, ‘neural networks’,’semantic analysis’, ‘natural languageprocessing’, ‘wisdom of the masses’, i.e.,can’t make a good omelet from a lot ofrotten eggs!Now I remember: Some people don’t likelong posts! So, there’s no room toexplain here!Back to writing some Web pages!

    1. William Mougayar

      I hear you. It’s all under the rubric of serendipity, although it’s a guided serendipity, versus totally random.Nothing beats precise filtering, but it seems that nobody wants to spend time anymore configuring their filters, let alone the fact that you can’t even configure filters well on social sources. Instead, we’re spending more time following people, sharing stuff, etc…and if anything we’re staying at the “gross level” on the personalization scale, and we keep assuming that the smart algo’s will get better over time, and we are ecstatic when the software surfaces something we didn’t know about it. It’s the sad state of having abundant information, and poor attention spans.

      1. sigmaalgebra

        Okay, now we have started to ‘identify’the problem.> It’s all under the rubric ofserendipity,Now, mostly yes. For what I have in mind(derived, coded, waiting for me to write afew more Web pages for the UI front end),essentially serendipity has nothing to dowith it!> Nothing beats precise filtering,Yes, but for what’s so far on theInternet, for what we are talking abouthere, that’s the empty set. Basically thefilter needs to be looking for contentwith desired ‘meaning’, and that’s a holygrail problem.> nobody wants to spend time anymoreconfiguring their filters, let alone thefact that you can’t even configure filterswell on social sources.Of course not. The filters suck. The’crucial, core’ technology of the filterssucks. Routine software is not nearlysufficient for the crucial coretechnology. The computer science part isfrom routine down to trivial. But thecrucial part is just not programming orcomputer science; the prerequisites in theCS departments are like hen’s teeth.Why? As I outlined implicitly, it’s justcrucial to formulate the problemappropriately. I mentioned several waysto make mistakes in the formulation. Witha poor formulation, get a ‘filtering’problem can’t actually solve well in anysense, and not effectively in practice.Such an appropriate ‘formulation’ has toconnect with the ‘business model’, etc.Lots of ways to make mistakes.As I outlined, the filters take in data,manipulate that data, and report results,and the manipulations are necessarilymathematically something. I gave a longlist of topics ‘intuitive heuristics’,…, ‘wisdom of the masses’ not sufficientto make a good filter.There’s only one way for me to contributeto a solution: Be what I am, a sole, solofounder. Right: Going for 1+ billionusers for 2/3rds of the content of theInternet. We’re talking what, here,ballpark $400 billion? Curiously thatdoesn’t make the math or computing anymore or less difficult.If I tried to explain how to solve thisproblem to Yahoo, Google, Microsoft,Time-Warner, AOL, Liberty Media (now,there’s an outside chance from JohnsHopkins!), etc., all I’d get would besilence or laughs. Why? There’s a reasonthe problem is not solved; nearly no oneknows how to solve it and is trying. Andlikely and apparently there’s not a CEO ofa big company anywhere in the world whowould know even how to direct anevaluation of my work. All I can do is dothe rest of the software and go live.For more, the key is the math, and math islikely and apparently not mentioned as aninterest on any VC Web site in thecountry! That doesn’t mean I’m wrong!Indeed, as has been commonly, andcorrectly, observed, being the only onegoing that way is from an advantage tocrucial (i.e., nearly necessary, notnearly sufficient). Further, I have yearsand stacks of books behind me that say I’mright. It’s a math problem, guys, and notcollege math, either; suck it up and getused to it. Not a math major in college?I understand: Tough to BS the tests, andsome of the single exercises could take aweekend.Some of the prerequisite math I’m using isjust beyond belief; tough to believe itcould be true; tougher to guess it’s true;tougher to prove it’s true; but it’s true,and the proof is rock solid. Withoutknowing the math, no one would ever guesssuch a thing; nearly no one who studiedthe math is interested in computing,entrepreneurship, or the problem, muchless all three.In the meanwhile, I get some justfantastic music to listen to as I type insimple Visual Basic .NET, ASP.NET, classinstance ‘serializable’ de/serialization(nice stuff), ADO.NET, T-SQL, HTML, CSS,etc.!E.g., I just found and ordered from Amazona DVD video of Lohengrin! So I get tolisten to a white swan pulling a goldchain pulling a small boat with a knightof the holy grail in shining armor aboutto banish the bad guys and the evil witchand save and marry a gorgeous woman! Withgorgeous music every second! Can’t beatthat!Wish I could be 15 again and sit with myfirst girlfriend, 13, sweet, naturalblond, overwhelmingly, breathtakingly,heart stoppingly, world class drop deadgorgeous (pretty girl!), by a wide marginthe prettiest girl in her graduatingclass, I still remember, go over the storyof Lohengrin, and watch the video on DVD,complete with the fantasy and Wagner’swedding music to get her all romantic!I remember like yesterday taking her toHoward Johnson’s for Swiss ChocolateAlmond ice cream!Think of her along with the music and getanother two pages of code written! Swap$399 billion of the whole $400 billion forher? Heck yes!Or if she goes to a Christian church,start with the “Pilgrim’s Chorus” from’Tannhรคuser’ and the prelude to’Parsifal’!Nerd me, I totally blew it or she wouldhave done the last two years of highschool with my diamond ring on her lefthand.Teenagers today don’t know how good theyhave it! The DVD for the whole 200minutes was less than $30!Just got a message from Amazon — the DVDhas been shipped!

        1. Matt A. Myers

          I become more and more curious as to who you are with every beautiful and insightful comment you make.Is there a way to find out more about you?

          1. sigmaalgebra

            Ask Fred. I’ll have to become a publicperson when my software is done.

          2. Matt A. Myers

            Okay. ๐Ÿ™‚

  17. Aaron Klein

    This is going to be a difficult feature to wean yourself off of on a Ubuntu Phone. ๐Ÿ˜‰