Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I built a $5k Raspberry Pi server (yes, it's ridiculous)

jeffgeerling.com

Post reply on HN
Print view View original

141–150 of 199 posts

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#141

Post by willis936 »

dekerta wrote: post #60
jjk166 wrote: post #53

Earlier quoted context omitted.

I honestly wonder which came first, the viewer preference or the algorithm. Like are there people out there that see a whacky face in the thumbnail and think "Oh that's a video I want to watch!" and the algorithm just got trained on that? Or did whacky faces just resemble something (from the algorithm's perspective) that was briefly popular and then a positive feedback loop of algorithm recommends it -> gets popular…

I think people's eyes are just naturally drawn to faces, so more people end up seeing the thumbnail among the grid of other videos, and are more likely to watch the it as a result

That's the base narrative, but I find it difficult to believe without proper controlled testing. People are 25% more likely to click on a technical video when the creator is trying to look like a buffoon? It just doesn't sound correct.

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#142

Post by pwnna »

geerlingguy wrote: post #94
ASalazarMX wrote: post #76

Earlier quoted context omitted.

> an exaggerated face, a bit more than that Have you tested a difference between a regular expression (intrigued, skeptic, smiling...) and a goofy face? For science?

Yes. Goofy face always wins :( Though it gets an insane amount of backlash on HN (case in point: this thread).

Maybe both your observation and HN is right: the distribution is multi-modal in the sense that (presumably) there is a large majority of people where the thumbnail works for and views the content, and then there is a small (and possibly vocal) minority of people who absolutely hate it and treat it as an anti-signal. I know I belong in the latter group and will lose all interest in the video once I see the weird face and the edited icon. I've seen a few of your videos and they're indeed very good. That said, I've noticed that I semi-consciously would avoid your video due to the icon which triggers an immediate reaction from me to stay away. It's a shame, because I know that the videos can be good, but the thumbnail is actively driving me away. In a way, I'm no different than the people who clicks on these videos, as I am also judging a book by its covers.

This kind of multi-modal behaviour can be very hard to capture with A/B test. You get a large influx of people that more than offsets the loss in the other audience pool. It's also anyone's guess on which group is better. From a pure business perspective, it's probably okay to lose me, as I use adblock and/or youtube-dl to watch the videos, which means I don't really drive revenue. That said, I've noticed some second order effects: as I'm more technical, I would recommend these type of content to others who will listen to my advice. I've noticed similar effects with others. If the more technical people are in the second group and are driven away, I wonder if this kind of stuff makes a difference with the long term health of a channel/content generator, which could be very difficult to quantify.

In any case thanks for your videos and insights.

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#143

Post by drcongo »

drumttocs8 wrote: post #84
rym_ wrote: post #7

I get so frustrated with the idiotic thumbnails on youtube video these days. I am sure it's all optimized for the maximum number of clicks/views but if I see a still that looks the way it does at the bottom of this page I generally just don't watch it on principle.

We see a face with an (albeit fake) emotional reaction, our monkey brains say "that must be interesting!" It's very stupid, but not surprising that those videos get more engagement.

My emotional reaction to those is "this must be shit, they've had to put together this awful Clickhole picture just to try to make people watch it".

I've literally never watched a single one.

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#144

Post by vmception »

dark-star wrote: post #126
WastingMyTime89 wrote: post #86

Earlier quoted context omitted.

> The question not being asked is how your audience changes A vs B in other ways than size. That's not how it works. The most important factor to get views on YouTube is the recommendation algorithm. It makes a huge difference. YouTube will not even push your videos to people who are subscribed to you if the recommendation algorithm doesn't favor them. As the click-through rate is an important part of how the algorit…

Youtube algorithms are truly weird. I have a YT account with, I think, 3 or 4 subscribers and maybe a dozen or two views on the 10-20 videos I ever uploaded. I basically use YT to upload things that I want to share with some close friends. Then, earlier this year, one of my videos suddenly skyrocketed[1]. It was an old video, 10 years or so, and it only consisted of one 10 second scene of a movie that I found funny a…

the only culture I like about Youtube is how the other commenters and viewers know that they all found each other due to the algorithm, because they watched anyway!

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#145

Post by stocknoob »

geerlingguy wrote: post #24
rym_ wrote: post #7

I get so frustrated with the idiotic thumbnails on youtube video these days. I am sure it's all optimized for the maximum number of clicks/views but if I see a still that looks the way it does at the bottom of this page I generally just don't watch it on principle.

Believe me, I had a nice discussion yesterday with my sister (who helped me take pictures for the thumbnail), and we both agreed it's one of the dumbest things about current YouTube... The thing is, from all my own A/B testing, a face vs. no face will automatically get 20-30% more views (all else equal). And an exaggerated face, a bit more than that.

What do you think the A/B test for showing cleavage in your thumbnails?

https://blog.codinghorror.com/how-not-to-advertise-on-the-in...

Where does it end?

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#146

Post by ryandrake »

geerlingguy wrote: post #64
foodstances wrote: post #54

Earlier quoted context omitted.

But why is view count the only metric you care about? How about getting fewer views but more engagement, with viewers that want to see the content but are turned off by the stupid thumbnails?

Getting fewer views (and likely much reduced watch time) means YouTube's algorithm quickly buries your video. You and I might use subscriptions on YouTube (just like I prefer RSS), but 99% of viewers for some reason just go to the home page and click on things YouTube recommends (regardless of whether they subscribe to a channel). If you don't get recommended, you don't get views, and it's even less likely that regul…

> Getting fewer views (and likely much reduced watch time) means YouTube's algorithm quickly buries your video.

Question remains: so what? Unless you are monetizing your videos and intend to make significant money from “engagement”, why would someone single-minded let optimize for views or algorithm-friendliness?

For this guy, I guess he’s trying to make a living from YouTube, so I guess… by all means growth hack your video thumbnail. But for the rest of the video uploaders, absent a desire to monetize, a video with 100 views is no worse for the uploader than a video with 100,000 views.

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#147

Post by riobard »

szszrk wrote: post #33
MisterTea wrote: post #19

Booo. Its just a pi with expensive ssd disks attached. Nothing interesting here.

Disagree. Plain Pi won't be able to attach 4 sata drives natively and is not a real alternative for actual NAS. This might be. Jeff does a lot of Pi-NAS testing and I'm thankful for that, as I find of the shelf solutions lacking a lot. Mostly software-wise. There is a huge demand for a simple NAS device that is not much larger then drives it uses, but doesn't have Qnap OS installed on it. It has good network connecti…

> a simple NAS device that is not much larger then drives it uses, but doesn't have Qnap OS installed on it.

Well, you could just install your favorite Linux distro on a QNAP NAS device. Cost-wise the Pi-NAS option isn't even competitive if you aim for similar spec and form factor and packaging.

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#148

Post by geerlingguy »

MisterTea wrote: post #77
geerlingguy wrote: post #34

Earlier quoted context omitted.

The difference is the Radxa Taco board, which interfaces a CM4 directly with 5x SATA III slots, 1x native NVMe M.2 (2280), and 1x 2.5G NIC. Most typical Pi 'server' builds slap all that IO on top of the Pi 4 model B's USB 3.0 bus.

And the Pi4 SoC has only 1 pcie 2.0 lane so there's little to gain over USB 3.0 save for some latency.

In my testing over the past year, almost everything gets a 5-20% speed boost by cutting out the USB 3.0 layer, except for large file sequential copies. It adds up, and is also more reliable and practical for things like ZFS.

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#149

Post by AnIdiotOnTheNet »

CyberRabbi wrote: post #140
AnIdiotOnTheNet wrote: post #127

Earlier quoted context omitted.

> The existence of systemd and chrome have unequivocally made the world a better place. Whether or not that is true is besides the point. The point is that it is valid to criticise them regardless of the fact that these things were given away for free. > You sound like someone who has never made anything independently in their life. Incorrect. > He has chosen to make a living by providing free technical content onlin…

>> The existence of systemd and chrome have unequivocally made the world a better place. > Whether or not that is true is besides the point. The point is that it is valid to criticise them regardless of the fact that these things were given away for free. You’re moving the goal post. My initial comment wasn’t because you criticized him. You may do that all you like without any issue from me. My comment was responding…

> You’re joking right? Free as in people do not have to pay money to watch his content.

Sure, in a very literal interpretation of the word "free", the same interpretation that would allow me to say "I'm free to not pay my taxes".

> You accused him of making the world a worse place for styling his YouTube thumbnail in a way that maximizes the number of people who click on his video. That’s obviously silly when his videos are adding much more to the lives of many hundreds if not thousands of people.

Can you not see how much this reads like marketing copy? You make it almost sound like they're doing the world a service by getting them to click on the video (thus generating money for them) they otherwise wouldn't have.

Putting that aside, as I mentioned elsewhere I am not convinced by the "ends justify the means" argument on display here.

> That’s like saying artists who make more appealing art are engaging in psychological manipulation.

They are, but that's also precisely what they're selling. We view art with the expectation and desire for it to manipulate us. I suppose you could make the argument that at least some people actually want that from youtube thumbnails, but I would point out that that is roughly equivalent to saying people actually like the addictive qualities of lootboxes.

Re: I built a $5k Raspberry Pi server (yes, it's ridiculous)

#150

Post by whywhywhywhy »

jrockway wrote: post #47
rym_ wrote: post #7

I get so frustrated with the idiotic thumbnails on youtube video these days. I am sure it's all optimized for the maximum number of clicks/views but if I see a still that looks the way it does at the bottom of this page I generally just don't watch it on principle.

I'm not convinced that it's optimized in any way. Linus claims to have A/B tested it, but changing the thumbnails disturbs the experiment too much (different time of day, Youtube knows that you changed the thumbnail and probably uses that information for something, CDN lag, who knows what else), and the sample size is too small to be statistically meaningful. I watch a ton of YouTube and exaggerated facial expression…

I'd believe it works in an A/B test, but the end of the day the reason it's higher is young children click on anything with that stupid looking face on it.

So I bet it does result in higher views, just children viewing it.

Doesn't mean you should do it though, maybe try having some dignity.

Post reply on HN