Hacker News Reader

Comment by c7DJTLrn | Hacker News Reader

Machine learning probably had something to do with it. It notices videos with thumbnails of people with their mouths and eyes wide open doing some kind of shocked expression with red arrows pointing to something perform better than average. Then it starts recommending videos with thumbnails like this preferentially. YouTubers take notice and start producing thumbnails like this for every video.

Go to the front page right now with clean cookies and you'll see most of the thumbnails have "reaction faces" in them for no apparent reason, usually pulling some kind of exaggerated expression.

Clickbait used to mean that you baited the viewer into clicking, now it means you baited the algorithm into recommending you.

View on HN

There was a Linus Tech Tips video where he admits that it's stupid, but sadly it works really well. It's a bit like when Wikipedia kept using Jimmy Wales' face in banners requesting donations. People claim to not like it, yet it's was what got the highest conversion rates.

Gawd those haunting eyes... I wonder how many people donated just to make them stop.

This is a clear example of penny-wise but pound-foolish. At no point in a product's lifetime does should machine learning obviate the need for good taste.

I'm not going to argue that, but that ship sailed back in ~2014. All the FAANGs do live A/B testing like this.

2014 lines up with my memory of when YT recommendations, among many things, started to change for the worse.

On a side note, wow. I can’t believe that 8 years have already passed. It feels like yesterday that people were actively debating these topics here on HN.

Patrick McKenzie (aka patio11 here) had a post/talk about how when he was working with FogBugz they realized that people LOVE to watch Joel Spolsky talk.

In fact, A/B testing showed that the longer the video they put on the home page, the higher the conversion rate.

I've long suspected many large production recommendation algorithms suffer from a positive feedback loop problem. In the example you point out, there probably is some real behavioral bias to focus attention on highly expressive faces. Then the algorithm picks up on this small bias and amplifies it, where the entire front page turns into young people with shocked faces.

When I was playing WoW, it was hard to find, for example, "guide to blackwing lair". It would be way down the page, below videos always titled "shocked reaction to Blackwing lair boss kill", "streamer reacts to BWL speedrun".

I think that’s an issue with the monetisation. People who needs a specific guide would be relatively small in numbers. Unless your video is watched by tens/hundreds of thousands, it is not worth making (unless for hobby). Content creators are feeling like they need to make the title and image appealing to a wide audience rather than people looking for something specific.

The search engine should prioritise videos matching your specific search so you can find what you are looking for from hobbyist creators but I’m not sure if that’s happening.

That shit is the fucking worst. 10 results down it gives you the "People also watched" and "For You" categories before it resumes listing what you fucking searched for. Can you imagine 20 years ago trying to build a search engine and explaining that you should insert search results that people didn't fucking search for?

Yes. Canva even has templates for clickbait, including "passive income" scams which have taken over YouTube.

https://www.canva.com/p/templates/EAFBCRpOy3g-yellow-black-e...

Wow nail on the head. Just did this and out of the first 8 videos, 5 were reaction videos, 4 with the exact same "shocked" faces. Scrolling down the page it is much the same. How can google allow whatever algorithm to present such homogenous content on such a general platform?

Notably though, if I log in and visit my home page I have only one video of this nature on the front page in the first ~50, so it does appear possible to teach the algorithm you will not engage with such content even without conscious effort.

Reply on news.ycombinator.com