Spells and incantations
This week: AI prompts, panopticons, GIFs and water machines
Welcome to this week’s free edition of The Terminal. If you’d like to become a paying subscriber and get access to more content, hit the button below.
Some housekeeping: I’m moving this weekly free instalment of The Terminal to a Monday, thanks to some new projects I’m working on. (More on that later.)
I co-host the podcast Down Round, which you can subscribe to here.
If you wanted to take a sacralised, enchanted view of the form and function of modern tech, you could formulate it like this: microchips are runic inscriptions on sand which, when coursed through with lightning, act as conduits for incantations that are written in occult tongues and studied obsessively by monks. (Imagining the average software dev as a monk might be a stretch, but let’s roll with it.)
But that mildly annoying stoner thought bubble is increasingly becoming a broader mental model for the industry and people trying to understand new shifts. As focus moves towards AI models that can accept and process natural language prompts in increasingly sophisticated ways, we’ve quickly arrived at a position where we can type something in human language and have neural networks conjure images, text and actions seemingly from thin air.
One of my favourite second-order effects of this shift is how contested the space of prompt design (or spellcrafting, if you like) is rapidly becoming. When AI image generators hit the market en masse, one of the immediate assumptions was that entire creative industries could be replaced by laypeople entering extremely simple text prompts. But one of the wonderful traits of humankind is our propensity to create replicable systems, and another is our ability to turn things that should be fairly simple into shockingly creative endeavours that require more skill than strictly necessary. (See for example the guys who have turned Google Maps into a high-intensity spectator sport.)
The still-nascent AI art space has quickly been colonised by obsessive pro users who have figured out all sorts of complicated prompt tricks to generate exactly the sort of images they want. This isn’t remotely comparable to the skill required to create a digital artwork manually — at least, I don’t think it is — but it’s some order of magnitude more complex than just typing “cyberpunk Shrek” into DALL-E 2. For example, here’s an image generated with Stable Diffusion:
And here’s the prompt used to generate it:
ancient indonesia, indonesian villagers, punakawan warriors and priests, cinematic, detailed, atmospheric, epic, concept art, wimmelbilder, matte painting, background mountains, shafts of lighting, mist,, photo – realistic, concept art,, volumetric light, cinematic epic + rule of thirds | 3 5 mm, 8 k, corona render, movie concept art, octane render, cinematic, trending on artstation, movie concept art, cinematic composition, ultra – detailed, realistic, hyper – realistic, volumetric lighting, 8 k
This isn’t really natural language at this point — its a series of vaguely linked incantations to coax a desired image out of the machine. (It’s its own kind of coding, in a way.) You can see why people are imagining the space will become even more absurdly professionalised:
But as quickly as AI art mavens figured out ways to systematise production, people started automating it. A number of prompt generators have sprung up in recent weeks, which use AI to create prompts for you, which you can then pop into one of the many image generation models to get better results. Amusingly, this has led to an immature kind of gatekeeping, as people who became proficient at prompt writing over the past few months become annoyed at being automated out of their newly-found professions.
Vale to the AI spellcrafters. Your reign as priests of the new order was brief and went largely unnoticed.
Meta has been attempting to acquire GIF-hosting platform Giphy since 2020 and integrate it into its platforms, and has faced serious antitrust pressure for it. Unfortunately for both parties, animated GIFs have become deeply uncool in the interim. Now, the increasing uncoolness of Giphy’s product is becoming part of the company’s desperate pitch to competition regulators. From The Guardian:
In a filing with the Competition and Markets Authority, Giphy argued that there was simply no company other than Meta that would buy it.
Its valuation is down by $200m from its peak in 2016 and, more importantly, its core offering shows signs of going out of fashion. “There are indications of an overall decline in gif use,” the company said in its filing, “due to a general waning of user and content partner interest in gifs.
“They have fallen out of fashion as a content form, with younger users in particular describing gifs as ‘for boomers’ and ‘cringe’.”
I’ve always abstractly been a fan of the animated GIF as a format. (Less so in actual execution.) They’re relicts from an earlier internet era — extremely inefficient, clunky and low-resolution imports from the Age of Forums which persisted for their organic cultural resonance rather than any objective technical benefit. I liked that, and I liked how platforms like Twitter had to develop hacky ways of making them work in a modern web environment, simply because everyone loved the specific aesthetic vibe of these busted, pre-Web 2.0 anachronisms.
Now the humble GIF is disrespected and spat on by Zoomers, merely because it has become an extremely cheugy transmission vector for the lamest aspects of 2010s mass culture. (Maybe we should have seen the writing on the wall when a flurry of thinkpieces in 2017 accused the format of enabling ‘digital blackface’.) All that said, if it means I never have to see the one of the woman spitting out water, I’m happy to lend my sword to their scorched earth campaign.
At Garbage Day, Ryan Broderick argues that GIFs fell from cultural grace largely because of big content warehouses like Giphy, which made them less anarchic, more searchable, more monetisable and less cool. “It’s sad and, I think, a very good lesson about what happens when you centralise a piece of internet culture,” he writes. “It becomes commodified, it becomes stagnant, and, ultimately, becomes cringe.”
A somewhat alarming project here from Belgian artist Dries Depoorter, who describes himself as tackling themes surrounding “privacy, artificial intelligence, surveillance and social media.”
For The Follower, Depoorter recorded a series of open camera feeds, then scraped the Instagram location tags for photos in the area of each camera. Using software, the two are linked, displaying the photos at their moment of conception.
The point he’s making is not that hard to figure out. But there’s an element to it Open video camera feeds remain one of the most fascinating artefacts of the digital age, and one of the few kinds of content which still feels somewhat transgressive to view. Yes, Deeporter’s project about surveillance, but part of what makes it so unsettling is the machine’s eye view of an unsecured video camera, which appears even more alien when juxtaposed with the performatively human lens of Instagram.
It’s funny how long we’ve been wrestling with that specific weirdness, too. In Don DeLillo’s 2001 novel The Body Artist, the main character spends hours of her day staring at a webcam feed of a road on the outside of the Finnish city of Kotka. From the text:
She set aside time every day for the webcam at Kotka. She didn't know the meaning of this feed but took it as an act of floating poetry. It was best in the dead times. It emptied her mind and made her feel the deep silence of other places, the mystery of seeing over the world to a place stripped of everything but a road that approaches and recedes, both realities occurring at once, and the numbers changed in the digital display with an odd and hollow urgency, the seconds advancing toward the minute, the minutes climbing hourward, and she sat and watched, waiting for a car to take fleeting shape on the roadway.
As one might expect, not everyone captured by Depoorter’s commentary on the overlapping systems and structures of surveillance society is pleased. From Input:
At least one of the project’s subjects is extremely unhappy. The Follower also identified Renata Costa and her partner Cleison kissing outside the Temple Bar. When contacted, Cleison declined to talk via Instagram video call, believing Input was behind the project. “I would like you to delete my photo... immediately,” Cleison writes. “It’s a crime to use the image of a person without permission.” Cleison says he’ll be contacting his lawyers.
It’s a MONIAC (Monetary National Income Analogue Computer), designed by New Zealand economist Bill Phillips in 1949. Originally intended as a teaching aid to help students learn how the economic processes of the United Kingdom worked, it used the flow of coloured water to represent money moving through the economy into various tanks, representing how it could be distributed and spent, and the effects of doing so. It was effective enough at a time where electronic computers were rare and not fully suited to the task of complex economic modelling that it was used for actual work and analysis.
Obviously, computers quickly got good enough at that work that we transcended our need for water-powered computers. Still, it’s fun to imagine a kind of quasi-steampunk alt future where everything runs on flowing water and fluidic logic. Fellow Vladimir Nabokov heads will recall that his 1969 novel Ada or Ardor is set in an alternate reality where electricity has largely been outlawed and everyone communicates with the ‘hydrophone’, which is some kind of (largely unexplained) telecommunications system which uses water to transmit voice.
Anyway. Here’s a fun video of a MONIAC in operation:
Fascinating read on unauthorised boutiques springing up in Mexico which buy, warehouse and resell clothes from Chinese e-retailer giant Shein, capitalising on the “lack of trust in digital businesses and low connectivity rates” across the country. (They also benefit from Shein’s discounts on bulk purchases.) E-commerce and its dynamic finds ways of manifesting even in those communities still resistant to it.
An interview with Craigslist founder Craig Newmark, discussing why the site still looks more or less the same as it did in the late nineties. “I've learned that people want stuff that is simple and fast and gets the job done. People don't need fancy stuff. Sometimes you just want to get through the day.”
Seems like despite Meta’s attempts to turn Instagram into a content firehose aren’t working very well. From the WSJ: “Instagram users cumulatively are spending 17.6 million hours a day watching Reels, less than one-tenth of the 197.8 million hours TikTok users spend each day on that platform.”
The New York Times paywall now uses machine learning to determine how many free articles a given user should get access to.
A fun summary at Jacobin of the unusual urban design and architecture obsessions of the freshly-minted King Charles.
On the demographic of remote workers most troubled by the push to return to the office: people who have secretly moved to another country.
A nascent mystery developing: who is Elly Conway, the debut novelist behind forthcoming spy thriller Argylle, currently being made into a $200 million Apple TV+ movie? She appears to have no internet presence, and the Hollywood Reporter made no progress trying to figure out her identity or anything else about her.
An essay on AI art: “We find ourselves currently somewhere between the uncanny valley and the holy mountain, alive in a time of dreaming machines.”
And here’s another front-end image generation interface for Stable Diffusion. Play around with it before it inevitably goes paid or becomes volume-restricted.
From 2019, but I found it interesting: The hunt for Japan’s ‘ghost wolves’.
One of those essays most people have read by now, but worth reposting given recent events: “Roger Federer as Religious Experience".