TITAA #30: Old Cities, Memory, and Lost Ships
Penis Chambers - Shrimptaurs - Caption Stories - Books about Memory - Ships Lost
Göbekli Tepe, the site of an unbelievable archaeological find in Turkey, is older by far than Stonehenge. Stonehenge’s simple rocks are dated to about 2500 BC. The Great Pyramid at Giza is 4500 years old. Göbekli Tepe is believed to date to 11000 BC, and unlike Stonehenge, the stone pillars feature lovely carved animals. Here’s a picture of one pillar from Klaus Schmidt, the archaeologist who worked on the excavation:
Sean Thomas was one of the first Western writers to visit and write about it, and his recent piece on the even older neighboring Karahan Tepe blew me away with its weirdness. It is thought to date between 11000 to 13000 BC. They have found repeated images of people with 6 fingers and evidence of a skull cult (defleshed skulls with holes for hanging them); “At a nearby, slightly younger site, the Skull Building of Cayonu, we know of altars drenched with human blood, probably from gory sacrifice.” The first chamber of excavated stones has been named the penis chamber, for its 11 giant erect stone penises, and everywhere there are carvings of men with their hands over their genitals. What the hell was going on here?
One sculpture found at the Karahantepe site that is even more fascinating than the rest is of a human figure carrying a leopard on his/her back. Prof. Karul told the Anadolu Agency that while it is hard to fully comprehend what this was trying to show, it did indicate that human–animal relations 11,000 years ago were very different. (source)
Uh huh. I can’t find a picture of that, damnit.
Göbekli Tepe was mysteriously buried by its creators in about 8000 BC, and evidently the even older Karahan Tepe was too. Karahan Tepe is enormous, and undergoing investigation now — it’s not a tourist site yet. “Archaeologists have excavated around 1% of the 60,000 sq. m site since 2019.” (source) “Karahantepe houses more than 250 T-shaped megaliths featuring animal depictions. Stelae with human depictions and three-dimensional human sculptures were also found in the site, which distinguishes it from Göbeklitepe.” (link) Around it in the “stone hills,” the Tas Tepeler, are clustered about 12 neolithic sites that have barely been touched yet.
Along with the mysterious carvings and stone temples, these sites have upended archaeological theories about why people stopped migrating and settled down in villages. It wasn’t because of agriculture, it seems: here, at least, it was for worship at their temple sites, with possibly bloody and drunken human sacrifice, and agriculture only came later out of necessity.
AI Art & Creativity Tools
Google released a DALL-E 2 competitor, Imagen, that they claim produces better output. Also no public access and “proprietary data” for the training. I’m not going to deep dive on this, especially since I don’t have access to either, but the results look pretty similar to the style of DALL-E 2 from what I can see on Twitter. Lucidrains has an implementation with some good explanation of how it’s much simpler than DALL-E 2 here. (No model, of course, but a call to action to recreate it with LAION.)
But I will deep-detour into the Shrimptaur. Jenni Polodna is a friend who is a very funny game writer and her shrimptaur generated on Midjourney went viral.
It was on the front page of Reddit, and random people I know say they’ve been shown it by their partners, by people in brew pubs, by feral squirrels in parks... Jenni says she has been followed by 1K+ people since the shrimp and Buzzfeed is doing a story on it. I’d like to add for the record that Midjourney can do fine cats too. (Truth in advertising, I am a mod there and really like it, as you know if you’ve been reading this nl; if you’re a reader friend I may have beta invites.)
DALL-E Flow, a human-in-the-loop text2image reviewing and generating process from Jina AI, is a nice experience, apart from the time it takes. But it is free. You get a grid of thumbnail options from 2 open source models (the in-training DALLE-E Mega and GLID 3 XL), and then pick which ones to refine with diffusion and upscaling. It’s free in a colab, but the experience of iterating and upscaling on Midjourney is much faster and often better quality. However, you do pay for that.
Majesty Diffusion is a recent combo release of some new latent diffusion work via @MultimodalArt. Their newsletter is doing a better job of being completist than I can be (without going crazy) about the latest fun models, open source datasets, and colabs for text2image, so do subscribe.
John Whitaker’s (@johnowhitaker) AI Art course with code notebooks now includes diffusion models and neural cellular automata. This is a wonderful resource. I need a vacation to work through it all.
Captions and Stories
Moving on: I really like the nascent creativity tools that support or combine both text and image generation… This is a gradio HuggingFace demo that combines simple GPT-J poetry with image generation using MultimodalArt’s Latent Diffusion app. It was built by @yvrjsharma.
Ok, the poetry is not usually very good. (I have worked on poetry generation and I have the expertise to say that.)
There is another text and image story generation gradio app here —which will also link your images into an interpolated video—but the outputs I’ve gotten are not fantastic. It’s harder to make a good story with GPT-J than it is to make a short snippet of poetic-sounding text.
More fun with caption generation, which I love inordinately for the errors: Socratic models does fun captions but has a hardish time with artwork.
Related, this MAGIC story generation is, well, a good idea but a bit meh? in their given examples. (Paper “Language Models Can See: Plugging Visual Controls in Text Generation” by Su et al.) There is a colab and also a replicate.com web demo. I got some strange results with a Midjourney image (below) I made from a Dwarf Fortress bug report: “Dwarf has mixed feelings about religion.” It went on about the dwarf being gay and I guess that’s why he was conflicted? When I retitled it to just say “Dwarf is Conflicted,” the results were still a bit weird and in this case, meat-focused and pro-vegetarianism:
Avatar CLIP - generating animated 3D avatars from text. There’s a colab and inspectable/downloadable 3D models, but you have to register and download some stuff (so I haven’t).
Learning Visual Styles from Audio-Visual Associations, a great idea with no code yet. Alter your image based on sound, like “rain.”
CLIP-CLOP: CLIP Guided Montage. Lots of people have linked to this (including me) but no one seems to be able to make anything good with it? Note there is a bug in the colab Arnheim 3 in which the wrong variable name is used for the patch set (change it from EXAMPLE_PATCH_SET to PATCH_SET). I really wanted to love this and if you made something good, send it.
Plantarium: A node-based 3d editor for generating plants by Max Richter, proving I can kill plants even on a screen in no time at all. (Code and demo.)
Videogrep: a tutorial to a tool for command-line editing of videos using the closed caption text, by Sam Lavigne. You can use regex’s, n-grams, etc. It’s pretty sweet looking.
Writing and Game-Related
An ACL workshop with paper links on Intelligent and Interactive Writing Assistants. There are some good papers in here, including one by Kreminski and Martens on unmet creativity support needs for creative writing helpers, song and poem generation, story writing. I haven’t had time to read them all.
Related: An interview with Ken Liu about writing a story with a text generator.
⭐ Lara Martin’s Storytelling (and game dev) Resources is a great list. Code, datasets, more.
Writing for Games, a new book by Hannah Nicklin. But, what, the Kindle version is $52? What is going on in publishing right now.
3D Math Primer for Graphics and Game Development, ebook now free.
Fab city generator shader exploration in an Observable notebook, with sliders.
Townscaper and the Design of Cities - a great YouTube video that walks through Townscaper in the context of urban planning, landscape, and architecture.
Babylon.js web rendering engine for game graphics.
NLP, Data, Infovis
How to implement Gradient Explanations for a HuggingFace Text Classification Model in Tensorflow by the ever wonderful Victor Dibia. With illustrations of the importance of word features to the classification.
Related, What Have Language Models Learned? A PAIR explorable from Google.
Dendromap: a tool (paper and code) to visualize (mis)classifications of large image datasets with treemaps. By Donald Bertucci et al.
Visualisation of text embeddings, structured, LMFingerprints, which I can’t seem to load right now but it looked really cool in the images and I like their other work.
The online visual Pandas tutor was ported to Pyodide. This means no server. W00t.
Related: Python in the browser, in pyscript, via Anaconda. So exciting!
⭐ Permutation City, by Greg Egan (sf). This kind of broke my brain: It’s about artificial life, embedded realities, memories decanted into copies in embedded realities, and the nature of self and perception. Also, obviously, about immortality and what that means. There are a lot of levels. Simplistically: An insurance salesman is selling virtual people further copies of themselves to decant into another embedded world copy, complete with an artificial simplified life universe inside it. You know, as insurance.
Severance, Ling Ma (lit/sf). I think I saw a rec for this from Mark Sample? It’s tough to describe. A virus has turned people into robotic zombies of their past selves, who mindlessly do what they used to do, make dinner, drive taxis, organize the store… Some few survived, including Candace Chen, who recounts her memories of being an immigrant, her parents’ struggles in a new country, and finally working her dumb job managing the printing of bibles via a Chinese factory. She stays at her job way too long, like the zombies, before getting caught up with a classic post-apoc-cult run by an asshole. But it’s mostly about being a zombie worker and memories.
The Memory Police, by Yōko Ogawa (lit/sf). Another one about memory, more viscerally disturbing. On a Japanese island, as the people lose memories of things in the world, the Memory Police require them to be destroyed. A few are immune, and must be hidden from the officials. A novelist is writing a novel, her editor hidden in a secret compartment in her house, and then novels are forgotten. It’s a difficult book.
The Man Who Spoke Snakish, by Andrus Kivirähk (fantasy). Another one about memory: In medieval Estonia, the last speaker of the animal language “snakish” lives in the woods with his family and animal friends. People keep leaving the forest for the villages, forgetting how to talk to animals, and their lives seem much harder afterward, especially compared to the forest dwellers who eat a lot more meat and less disgusting bread :). The forest isn’t entirely idyllic, as it includes a fanatic spirit worshipping priest who is a danger to all. The snake family and the primitive people who live in caves and trees are less crazy. It’s fun and funny and sad, although maybe a tad long.
A Master of Djinn, by P. Djèlí Clark (fantasy). A fun romp in an AU steampunk modern Egypt with djinn integrated into society. It was great to get kick-ass Egyptian women as the heroes and see their perceptions of the English colonial remnants still squatting and sneering at the locals.
Eyes of the Void, by Adrian Tchaikovsky (sf). Sequel to Shards of Earth. An excellent vacation read, I was glad to get back to this universe and these characters, and Idris seemed a bit less passive and whiny. Some good revelations and more Originator archaeology. However! Not the end of the series yet.
⭐ The Shining Girls, on Apple TV (sf), is based on a good book by Lauren Beukes. Impeccable casting and performances. It’s about a time traveling serial killer and a former victim (played by Elizabeth Moss) who survived and is now hunting him, as timelines change around them. She starts as a cowed, depressed character destroyed by what happened to her, who gradually wakes up and becomes angry and pro-active as she goes on the hunt. CW obviously: Women stalked and killed by man.
Signora Volpe (mystery) on Acorn TV (via Amazon). Emilia Fox (hah) stars as an MI5 agent burned out on her job; she moves to Italy and ends up using her spy skills to solve some local mysteries. It’s largely impeccable Umbrian scenery porn but was very fun (3 movie-length episodes). I’d like more, please. With a fruity red.
Disenchantment on Netflix (fantasy/comedy). This is the Matt Groening fantasy sitcom, which I am finally watching. It is very amusing, in a shitty world that needs cheering up. I love how the demon is seen as a cat by everyone.
Copies of Her (thriller) on Netflix was a good vacation thriller, but it did suffer from people doing dumb things, especially the daughter who discovers her mother has scary secrets. And the mother as a kid was a robotic doll who seemed completely unrelated to Toni Collette. I still enjoyed it on a holiday.
Ozark (Greek tragedy) is just a giant Greek tragedy now, or always was, and I’m not sure I can anymore.
⭐ Aniara (SF) a Swedish SF film I saw on Amazon or Netflix. A luxurious spaceship of immigrants leaving an eco-ruined earth to settle on Mars is thrown off course, heading out of the solar system. I watched this a few months ago, and it’s tough and sad, but I haven’t been able to stop thinking about it. On board is a semi-sentient AI program that summons up immersive images of earth nature to soothe people, in larger and larger desperate numbers. I keep thinking about the “art therapy” uses of text2image that I see.
It turns out this movie is based on a long 1956 sf poem by Harry Martinson.
For the fierceness of space does not exceed mankind’s. No, human cruelty stands up more than well. In the desolation of a death-camp cell space made of stone enclosed the souls of men, and the silence of the cold stones met the ear: Here mankind rules. Aniara’s ship is here.
⭐ I really enjoyed NORCO, which I mentioned before.
I played a bit of Cozy Grove, which is a very casual and light game where you just roam over a cute island looking for things to give to ghost bears. Plus some economy of trading and crafting. Incredibly un-demanding and just a “turn over all the stones again” kind of game.
I’m playing the Last Express now, a classic mystery on a train story game. I had to look up whether I could win in a knife fight (it’s about clicking correctly, you can). It actually reminds me a bit of Inkle’s Overboard.
Not a game at all — VR museum of war, in Ukraine. Experience bombed out remains of building sites in the Ukraine in VR.
The sun is lord of life and colour,
Blood of the rose and hyacinth,
Hair of the sea and forests,
Crown of the cornfields,
Body of the hills.
The moon is the harlot of Death,
Slaughterer of the sun,
Priestess and poisoner she goes
With all her silver flock of wandering souls,
Her chant of wailing waters,
The bed of shimmering dust from which she comes
Bound all around with bandages of mist….
The living are as blossoms and fruit on the tree,
The dead are as lilies and wind on the marshes;
The living are as cherries that bow to the morning
Beckoning to the loitering stranger,
The wind, to sing them his eerie ballads.
The dead are as frozen skeleton branches
Whereon the stillness perches like an owl….
The dead are as snow on the cherry orchard.
We have no easy months, do we. This one was a special clusterfuck for America and Americans. Not even to mention the ongoing Ukraine horror. The eternal search for escapism is eternal, while real life is so in our faces.
Hang in there, and say hi on Twitter,
Lynn / @arnicas