Discover more from Garbage Day
An endless world of boring robo art
Read to the end for a really good pangolin video
Here Comes The A.I. Malaise
Ammaar Reshi, a design manager from San Fransisco, is going viral right now (in the bad way) after he used a couple A.I. tools to create a children’s book in a weekend. Reshi used ChatGPT to build prompts for a picture book about how artificial intelligence works. Then he fed those prompts into Midjourney, settled on some characters and a somewhat-consistent aesthetic, and put it all together in a book. Then Reshi published it on Amazon Kindle Publishing and ChatGPT helped him fill out some of the details and metadata for the listing.
And, well, the reviews for the book are very bad. “The writing is stiff and has no voice whatsoever,” one user wrote. “And the art — wow — so bad it hurts. Tangents all over the place, strange fingers on every page, and inconsistencies to the point where it feels like these images are barely a step above random.”
I wrote about this in the Garbage Day weekend edition and I still think the experiment was kind of neat. Also, Reshi is far from the only person to try this. YouTube is now littered with examples of how to use A.I. to make thumbnails, motion graphics, comics, whole ebooks, and even motion-captured avatars. And it makes sense that the people who seem to be most excited by generative A.I. right now, aside from investors, are content creators. Those are the people who are under the biggest constant crunch to produce, are operating on the tightest budgets, with the smallest teams (if they have one), and have the least control over how their work is distributed.
And so right now, we’re seeing a glut of A.I. being created to populate feeds run by algorithms. Like Reshi using one A.I. to write prompts for another A.I. to make a book that is then submitted by an A.I. to Amazon’s algorithmic marketplace. It’s layers of automation on top of automation and I’m not surprised that final outcome is borderline gibberish.
The general reaction from Twitter users and artists around the web to generative-A.I. is very reminiscent of NFTs. It’s many of the same ardent Silicon Valley capitalists and grifters trying to commoditize “art” in ways that hurt real artists. And even more interesting, a lot of the same infrastructure used by NFT communities is now being used for developing A.I., like Unstable Diffusion, the pornographic A.I. Discord server with around 70,000 users that just launched a Kickstarter last week.
I won’t share screenshots from the server, but it works the way any other A.I. art server works. There’s a bot you can feed prompts to, which it uses to make images. Except this bot is built on the open source Stable Diffusion A.I. and is trained specifically to make porn. The result are dozens of chatrooms full of equally dream-like and horrifying pictures of genitalia and nude bodies slowly come into focus through community iteration.
The admins of the Discord asked for $25,000 to “train the new model with 75 million high quality images consisting of ~25 million anime and cosplay images, ~25 million artistic images from Artstation/DeviantArt/Behance, and ~25 million photographic pictures.” They hit that goal in less than 24 hours and are on track to double it before the week is up. The Kickstarter mentions the “democratization of art” a lot, but, notably, doesn’t mention where the 75 million images are coming from.
But the swift success of Unstable Diffusion’s Kickstarter means A.I. art has passed its first real hurdle towards large-scale mass adoption, one that NFTs and crypto never actually did: people really like using it for porn.
I get the impulse to treat A.I. art like the crypto craze all over again, but, as I wrote the other day, long-term, I think A.I. adoption will mirror smartphones more than anything else that’s come before it, and in the short-term, I think all of this feels more like search engine optimization more than anything.
SEO is an entire industry that popped up for humans to help companies talk to machines better. But generative-A.I. goes a step further because it’s not just a team of people helping you stuff robospeak into your blog posts. A.I. are now able to write the blog post, create the images that go with it, and even code the website it’s hosted on.
And it’s likely the content feedback loop will soon close and, while online content might be enjoyable to us, the internet will mostly become automations communicating with other automations. Even porn. It also seems likely that we are very soon going to reach a point where we start poisoning the well and generative A.I. starts influencing generative A.I. There are already a lot of A.I. art platforms that ban A.I. art from their datasets, but this is stuff is open source and ubiquitous and I doubt that can hold much longer.
So my guess is that one of two things happens (or both simultaneously). One is that generative A.I. of all kinds becomes both omnipresent and also creates a serious malaise. I think I’m already feeling it. I’ve noticed a certain vibe with A.I. content, like all the edges have been smoothed away, and I don’t hate it, but, perhaps worse, I’m getting bored fast. But I also think that as A.I. begins to eat its own tail we may end up seeing really weird and novel forms of content that humans really like, but can’t totally explain why. Like ASMR, slime videos, or that weird Spider-Man and Elsa YouTube channel from a while back that hypnotized children, it’s very possible the machines make something we can’t look away from.
I mean, I hate to even ask this question, but what if an A.I. invents a fetish?
This Is The Last Week Of Garbage Day For 2022!
I’ll be continuing the weekend edition for paying subscribers and the Discord will be running, but I’m going to take the next few weeks to do some extremely boring business admin and, somewhere in there, relax for a couple days. But keep an eye on your inbox, I might have some surprises for you before the year’s over. If you want to keep up with the world of online garbage, though, hit the green button below and subscribe!
What If This Is All Because Of Twitter Spam?
Elon Musk, and the crowd of venture capitalists he’s surrounded himself with, are really obsessed with spam on Twitter. Which has been something of a head-scratcher for a lot of folks outside of their clique, myself included. I’ve used the site every day practically for 15 years and have had over 50,000 followers on the platform for almost five years now. And the amount of spam I was encountering on the app was maybe a couple of weird tweets or DMs a month, if that — as long as I didn’t mention crypto.
Anytime I tweeted about crypto, my replies and other available channels would fill up garbage. And this was even more pronounced if I ever entered a Twitter Space. The other day I gained 300 followers after a Twitter Space, and 200 of them disappeared the next day. My DMs were, and still are, completely bricked with spam. It was a mess!
I struggle to take anything Musk says seriously, but I do wonder exactly how much his view of Twitter has been colored by the fact his mentions are absolutely crammed with crypto shills. This is, of course, very funny, because it’s his fault he’s associated with crypto in the first place. No one asked him to pump doge coin, but he craves attention from the worst possible people on Earth and, well, he got it, baby.
Musk’s big anti-spam feature is the Twitter subscription, which is apparently coming this week. The hope is that it could replace the advertisers that are leaving in droves because they don’t want their content sandwiched between a meme about Hitler and a 28-tweet thread accusing The Beatles of being satanic pedophiles. According to a new report from the Wall Street Journal, it looks like advertising is really dropping on the site. Which is probably going to be a problem because if Musk wanted to replace Twitter’s $5 billion-a-year ad business with an $8 monthly subscription, he would need about about 52 million users to pay, or about 10% of the site’s monthly active users. And I’m very doubtful that will happen, especially as Musk goes further off the rails and deeper into a culture war that feels completely stuck in 2016.
It’s as tragic as it is funny. Musk, desperate to be liked, shills doge coin, becomes synonymous with crypto scams, comes to believe that all of Twitter is crypto scams, breaks the site trying to fix his own dog shit mentions, and destroys a multi-billion-dollar ad business in the process. Or, as this meme much more concisely puts it:
Binance Isn’t Looking So Hot Right Now
Binance is one of the last big crypto exchanges standing after an absolutely crushing year for the market. But it’s also been under investigation by the U.S. Justice Department since 2018 for possibly breaking anti-money laundering laws, among other things. According to a Reuters story out today, some inside the Justice Department think there’s enough evidence to prosecute higher-ups at Binance, including CEO Changpeng Zhao. So that’s not, you know, ideal.
And, lastly, Binance locked withdrawals for certain accounts over the weekend. Oh, and it suspended a user who complained about it. Uh oh!
Anyways, if you’re curious what the latest is with FTX’s Sam Bankman-Fried, he was audibly playing a video game while participating in a Twitter Space today.
A Good Video
And it’s uploaded onto a platform where it’s far less likely to mysteriously disappear. If you missed the entire context here, Elon Musk was brought out on stage during a recent Dave Chappelle show and was booed mercilessly until he slinked off stage. I wonder if thinks these people are spam too.
I Really Think We Should Be Paying More Attention To Augmented Reality
I’ve been a long-time advocate of AR over VR just because I’m more interested in bringing the internet into real life than I am of bringing real life into the internet. And I really feel like we’re just one good AR wearable away from something really exciting happening in this space. Which I got curious about a couple months ago.
NREAL makes a pair of smart glasses that a lot of YouTubers have tested out. And there are obvious limitations — their goofy look and lack of a battery being the biggest hurdles — but I think there’s a lot of tech floating around right now that will pretty immediately lock into place once we can wear it on our face and project it into the world around us.
A Good A.I. Pic
Click through for the whole thing.
James Gunn Versus The Algorithmic Nerd Culture Industrial Complex
Filmmaker James Gunn, who is now one of the new heads helming the entire DC film division, got into it over the weekend with a leaker that goes by The Den Of Nerds. It was a fun Twitter spat and it’s nice that Gunn is as available as he is, but there’s also a few interesting fandom trends happening in the interaction that I want to go through.
First, the DC movie fandom is wildly divided. Ever since the Snyder Cut stuff, there has just been a near-constant fight between fans about what tone DC movies should have — bright and optimistic or grimdark with sludgy CGI. There’s also now an entire universe of YouTubers who create content to feed these kinds of fandom conflicts. Which is why The Den Of Nerds is tweeting stuff in the first place about Gunn hating Henry Cavill (Zach Synder’s Superman).
But, also, beyond just DC, in particularly male-dominated fandoms, “fan theories,” casting rumors, “leaks,” and fan fiction all swirl together and it’s gotten particularly bad around superhero movies this year. This fandom miasma really negatively impacted fan reaction to Doctor Strange In The Multiverse Of Madness after it didn’t end up delivering the cameo-fest that leakers were promising. And it has really gotten in the way of DC getting a handle on their own fans, who have engaged in Gamergate-like behavior not just towards each other, but stars of the movies, as well, like Aquaman’s Amber Heard.
And a lot of this cultural work has to fixed at the top so it’s slightly promising to see someone with the nerd bonafides that Gunn has being more vocal about this stuff.
There’s A Tumblr Bot Named Frank And She’s Surprisingly Smart
This is so cool. There’s a Tumblr chat bot called nostalgebraist-autoresponder that runs on the GPT-J language model and it can answer ask messages and reblog posts that tag it. The bot’s name is Frank and she doesn’t learn new words from Tumblr users that interact with her, but according to her FAQ, is learning from user engagement on her posts.
“She does learn from how many notes her posts get, helping her write better/funnier posts over time,” the FAQ reads.
The bot went viral over the weekend, so she’s lagging a bit at the moment, but I highly recommend checking out what she’s been posting.
Some Stray Links
P.S. here’s a really good pangolin video.
***Any typos in this email are on purpose actually***