YouTube Algorithm Basics (Cristos Goodrow, VP Engineering at Google) | AI Podcast Clips

50 thoughts on “YouTube Algorithm Basics (Cristos Goodrow, VP Engineering at Google) | AI Podcast Clips”

1. There is a Pinterest talk where they explain a very similar way to show related pins to the one you are seeing. I find fascinating the vectorial maps that emerge from this sorting. Meaning arises as footprint.

2. Useful. Before I didn't know how YouTube works. Thanks!

3. I'm loving the video clip. Thanks for this content!

4. Lex your podcast has been on a roll recently. 😀 keep educating us normies hahaha

5. Well, doesn’t the algorithm peep into the comments to learn more about the video?

6. It was funny hearing my 7 year old try and explain ai to her mum from listening to your channel lol

7. I'd love to be surprised by the algorithm…and less be spoon-feeded with similar content.

8. shadow banned

9. Basically describing how Google creates filter bubbles, fueling an increasing social divide, undermining our ability to get along with and talk to people with different opinions.

   It would be great if Lex did a podcast on this topic – how AI tends to reinforce human bias.

10. You mentioned in your video that you see channels with 57 views. Who are you talking about? I would love to see your thought process of who is interesting to you based on your personalized seeking device of youtube.

11. I've bee wishing for years that a user could just turn it off sometimes, I used to love the absolute random nature of YouTube, I would stumble across so much fascinating stuff that I would never have thought of looking for.

12. To help one's journey, the graph should have a concept of derivative, or momentum, which is representative of the user's changes of preference through time, and can be predictive of where the user should be.
    A graph is somehow reflective of the present related videos, built from other users that come in many directions of discovery and growth.
    But a person may want to see a video, today, that the algorithm may suggest in 1 year of slow discovery.
    So I'm suggesting an accelerated rate of discovery based on one's objectives and personal journey.

13. I feel like the YouTube algorithm quality has decreased in recent years. I no longer get steady, good recommendations for videos similar to those I’ve already watched. I feel like YouTube does a lot more curating of what it ‘wants me to watch’ (and excluding of what it doesn’t want me to watch) than before.

14. A real AI would watch the video for me and try out the recipe…

15. The best recommendations aren't in the search, they are more in the suggested videos under a particular video

16. One issue on youtube is the negative nudge. I want it to encourage me watching the harder stuff and discourage me watching the shitty stuffs, but so far it's been doing the opposite.

    Means I have to do more efforts to find quality videos that demands more effort to watch, and make no effort to find shitty stuffs (or make a deliberate effort to avoid it). For that matter it acts the same as TV.

    I would love Youtube to find people that want to be encouraged in certain directions (people making effort to find better content not clickbait or cheasy stuffs), and help them go that way.

17. My second recommendation on this video is a "30 minute timer", which is not particularly useful as I do not have any underseasoned chicken in the oven.

18. No solution to discover unknown quality content 😏😒😒

19. Yeah, tell us more about vectors in space.
    Then look at "Trending" tab and deliberately type "fox news" in the search. You'll get 4 out of 6 slots being occupied by CNN. Suddenly Google struggling to find what you are looking for.

20. Yes Lex! I subscribe in the same way & Sometimes I want to immerse myself into your way of thought.

21. So many shallow and disinformed videos talking about the YouTube algorithm and how to get more views, etc. And then, there is this. Straight from the "horse's mouth", on a quality channel, with quality people.

22. You know the recommendation algorithm works for “pro-democracy”, right?

23. For people looking to find out similarities in video taste with their mate, here's a little step in that direction – http://yt-set-ops.surge.sh/

    (Github – https://github.com/scientifichackers/yt-set-ops)

24. Lex Fridman Profile is at https://www.minds.com/newsfeed/1068582932929941504?referrer=coreychambers

25. Interesting to hear how other people see the subscribe button. I tend to watch people’s videos without subscribing but it’s mainly because I only like a specific series form a creator but I don’t care about all of their videos. I mostly subscribe to people that I want to see all of their videos. Does anyone else use it differently?

26. Rip Kobe

27. Great interview Lex (as usual). I’m now clear as to why the YouTube algo sucks. It’s gotten so bad that I want to get a job there to improve it. 🥴

28. What he is explaining, which sadly isn't much detailed, is a flawed system from what i can tell.

29. I only click 'like' to examine how the algorithm is affected. I only like a handful of videos at a time and use the algorithm as a tool for psychoanalysis.
    YouTube seems to lack a sense of communal responsibility. I want to be a mathematician someday, hence my time on YouTube should be spent watching lectures. YouTube seems to value trapping user's cognitive energies in a euphoric vortex rather than promoting responsible viewing habits.
    Users function as an alternate algorithm. Some users leave comments, some of these users make 'liked videos' public. To my mind this playlist functions as a set of recommended videos.

30. Did you guys know Lex is Khabib Nurmagomedov
    "s cousin ?

31. Interesting conversation the youtube Recommendation Algorithms kind of know what you're likely to watch but i'd love to be recommended some shorter videos like 5 mins and under also a bit more variation or new people it seems very conservative in what it recommends like they're so nervous to make a wrong suggestion that it keeps you in a bubble Take a risk sometimes to recommend someone unknown or slightly different than the topics that would make things more interesting .Interesting talk tho

32. Try to get Jim Simmons on here!

33. Lex is cool but I think he fakes his weirdness a bit.

34. the youtube algorithm recommended this. Google job Google.
    Can youtube's algorithm ask users who watch just a little bit of a video for feedback?
    Rosalid Picard mentioned that people speak implicitly to each other, but AIs can't understand implicit language. AIs need to be spoken to explicitly.
    For Cristos, i wish Google Play Store had a category for AI apps.
    i've noticed there's no category for AI apps.

35. Good insights. Was censorship and shadowbanning off the table as ground rules? For example, I still can't find the former mayor of New York's channel by searching 'rudy giuliani' – can anybody else see it in search? Scuttlebutt is that the only way in is with a direct link.

36. Is collaborative filtering similar to or enable stigmergy? I am familiar with the concept of stimergy (having an interest in ants and so on) and saw it as analogous to some phenomena on YouTube. I have noticed some bodies of users with particular interests on YouTube who appeared to achieve spontaneous order and function as a complex adaptive system. They appeared to be self-organising, self-contextualising, self-perpetuating and self-renewing. Each performed tasks according to their own personal autonomy. The vast majority appeared to only view content but some performed specific tasks (e.g. scouting new content to be introduced) and networked to one another. They often shared content in their feeds and playlists. However, even without users sharing content publically the system continued to persist. Also, these videos were only valued for visual aspects and often times there would be a mix of languages (e.g. English, Spanish, Portuguese, Russian) although videos would tend to mostly gravitate towards their own. But the suggestions were interestingly enough very well contextualised in predominantly gravitating towards specific visual themes. The interesting thing is that YouTube for many years only took action on the individual/local level upon reports within toxic systems on their platform. This meant they were enabling all sorts of criminals and dangerous groups on the systems level. Removing individual nodes from complex network is a futile effort.

37. Incredible content. Besides YouTube how would you get exposure to the YouTube recommendation algorithm.
    Hearing his prospect of users in space is awesome

38. Reading the comments. The algo is pretty good the only thing I would like as a user is to be able to have “user sessions”. I act as three or four different users, meaning that I have different intent (optimize recommendations ) and would like to differentiate session intents

39. Youtube should be owned by it's users and become a not for a profit public service.

40. Youtube should improve their ad recommendation system. I keep getting Etoro ads despite me not even google searching them even once in the past 6 months. All my searches are about python, C++ and coding. Yet, i get some stupid bootlegged online trading ad.

41. Hi Lex, I'll try give you a simple answer for your question. The reason why there're so many great Youtube videos with only 50 views is because few people think that is interesting. The best Youtube video about consciousness can be very insteresting to be, but unfortunately not for everyone. An average person likes superficial matters. I fully agree with you, but our view that the algorithm is often unfair in judging quality arises from the fact that we are not ordinary people. I created several channels on Youtube and it always bothered me that I had to create click bait content. But now I'm starting to realize that our measure of success is wrong. The measure of video's quality is not in its popularity, but in the specific people that video impacted. As a content creator, I can choose to work to reach 1M subscribers or just 100 subscribers. In this second case, it would be working to get 100 very interesting people who really know how to appreciate what I also appreciate. Quality is not in the numbers, it is in the people behind the numbers. In that sense I think Youtube has done a good job.

42. Watched till the end to help Lex's channel.

43. I’m regularly clearing my search end watch history

44. 8:05 I was wondering if we can find a partner with our youtube watching pattern. Better version of Tinder.

45. Algoritma basic kata dia? Ketidakmampuan dia dan banyak orang lain makanya SELALU MENGATAKAN BASIC. padahal kalau ditanya pada mereka apakah TAHU DAN APAKAH DAPAT MEMBUAT sebuah teknologi yang dapat menciptakan hal hal baru bahkan pemikiran baru level KING?

46. Only 16000 views but i am sure that most of it are from youtubers

47. Very much

48. Lex was a lot more humble back then

49. Leave the drugs. You are messing yourself.

50. Lex your podcast has been on a roll recently. 😀 keep educating us normies hahaha

Comments are closed.