I'm writing a second part of this post right now - about the differences in how separate online communities are drawing horses. There are clear differences in richness of detail, drawing mastery, but also in the amount of dicks drawn. My suspicion that people with more drawing experience draw less dicks simply because they already have drawn all the dicks previously. While people with an untrained drawing muscle perhaps fall back to these easy and familiar shapes. Drawing a horse is hard.
People draw horses because they want to draw horses.
People draw dicks because they feel it’s funnier, it’s a form of rebellion, and/or it’s way easier than almost anything else. And because they can without repercussion.
I understand all that, but how would you explain that 8,6% of drawings from people coming from HN were dicks, while visitors from Tumblr would draw dicks only in 1,6% of the cases? My guess is that people from Tumblr already explored a lot of the transgressions, so it does not interest them much - just like "potty talk phase" stops being interesting after your reach a certain age.
Let's call it unpaid software testing team :) People want to know how good is your filtering.
Anecdote time: Last year my wife asked ChatGPT to make a multiple choice test to learn some Biology topic. I arrived and she showed the test to me. The options were a, b, c or d. I wrote e to see what happens :) ChatGPT gave a good reply explaining that I made a mistake :(
Indeed, I think this is how it starts. However the filtering system does not indicate to the user if their drawing was filtered - otherwise it would turn into a "trick the system" game - a whole different genre. So the visitor always gets the satisfaction of seeing their penis galloping on the horizon. Still, many if not most penis artists then continue to add more and more lewd drawings. So it's not just a test but some other behaviour. Coming from Hacker News, they might feel like a hacker, a rebel breaking the system. And that's fine - after all there are not many ways in a society where you can feel this way and not hurt anyone.
It's the other way around - I integrated some things I wrote in the comments here into the blog post. I'm still writing the post though - it's not easy to hypothesize about people's intention without sounding like an ass.
So many dicks! Roughly 4 percent of all drawings are dicks. With over 600 000 drawings that means.. Twenty five thousand dicks.
I thought about including them, but as with all the other transgressive drawings, the only person who enjoys looking at them is the one who drew them.
Where is the generative dick doodle model? Consider the amount of manpower that could be saved by automating this labor-- probably equivalent so saving thousands of lives per year!
Oooh, there are just so many horses - more than 600 000 - that it's a tiny chance to actually see your drawings! And there are around 50 000 drawings marked as non-horses - all kind of weirdness - that could be an exhibition on its own..
Thank you for drawing, a typical drawing is seen more than a hundred times, so your pegasus and space invader were seen and appreciated : )
Nice game. I draw like 10 "horses". The first didn't move the legs because I didn't understand the idea, then I started to add moving things, and then weird things like the space invader. Anyway, mines were not as good as other examples I saw.
Yes, on the original gradient.horse it was an early experiment in hidden moderation - and it did help hide quite a few of the less kind drawings. Now, as the moderation model trained on the larger corpus works quite well, it's just for fun.
Would that be enforceable? Can people be excluded from the right to a legal dispute? It seems that otherwise any contract would include a clause that it can not be contested - which it sometimes does with mandatory mediation.
I think Mozilla is still mostly made up of tech-optimist people, so they were open and interested in ai from day one. I highly doubt there was any malicious intent.
Last I heard the ads were introduced to be less dependent on Google money - they actually cover the costs/salaries of the internal MDN team and thus secure the existence of MDN within the organisation.
Also apparently they are non-tracking ads, and so provide only a small fraction of income that tracking ads would bring, but that would go against the ideals of Mozilla.
So I'm seeing the ads as a net positive. (And am surprised that the people visiting MDN don't use an ad blocker anyhow).
I support MDN by disabling UBO. This is not (much) of a burden: there is no animation, and the advertised products are always technical and possibly of interest. Only complaint is the occasional large light areas, which do not play well with dark mode.
