TOC with links:
3 Great Museum Exhibitions in Paris
I am running late with your links because I was in Paris for work, and spent the last 2 days running through 4 exhibitions in our horrible early heat wave. I’m still recovering. But I wanted to push my 3 favs at you — the Hilma af Klint “Paintings of the Temple” in the Grand Palais is really good: the canvases are enormous, so it’s a bit awe-inspiring. They don’t shy away from her esotericism, with discussion of her automatic writing, seances, theosophical connections, invented language/scripts, and gender focus. Here’s an inexpert collage of some grabbed at random from my camera roll:
For companion in weirdness, you cannot beat the Leonora Carrington exhibition in the Musée de Luxembourg (the gallery just north of the park, which is by the way 5 degrees cooler than surrounding streets, at least in the shade bits). There is a lack of English in the bookstore, which is sad for many, but they have some good merch. Again, they go hard on her interest in magic and occultism. Very good coverage including childhood art (astounding). Here is a collage of details from some:
Finally, the Cartes Imaginaire exhibition at Bibliothèque Nationale (François Mitterand, main site) has super historical maps with many sea monsters among other decorations, plus literary and artistic maps. Even games!
Onto the links… a bit sketchy! I’ll add the TOC on the web version.
Google Flow has added Omni, a video-to-video editing model, which can also be used via a new agent tool (everyone is adding chat agents to their pipelines). Image-to-video coming soon. I should be celebrating, right? Welp: I was planning to show off some video I took in Paris and converted to 1900s Paris, but every.single.thing I try in Omni, even with Veo generated video, is blocked by unclear, mysterious “safety policies.” I’ve filed feedback a LOT. It’s impossible to use. ☹️ [Someone on X thinks it’s a country barrier, but France is listed as ok, so—? ETA: On VPN I got further with the Gemini App, so maybe it is. But not far enough to be worth showing it off.]
Two mentions of AI video at Cannes of interest to me — Kling AI Debuts at Cannes — Technical Discussion on Born of the Tide (with video we can see) and then “Goodnight Lamby,” credited to artist Dustin Yellin as film-maker with Darren Aronofsky/Primordial Soup as producer in “Cannes Classics” (and starring Paul Rudd and Chris Rock?!); but elsewhere discussed as a Google Deepmind collaboration for Aronofsky (link). There’s a greater in-depth article here in Indiewire with a tiny teaser. I think lots of classic fx went into this too, not just Veo.
Chloe VS History (via Kevin Kelly) — YouTube AI history channel that Kevin Kelly recs. Because influencer style educational history videos, why not!
Justine Moore on X wrote up her observations from Amazon’s aionthelot — re Hollywood interest in AI. “Studios are hunting for AI talent.” They should hire the guy who made the Michael Scott introducing Karpathy to the Office video (x). (Karpathy joined Anthropic.) OTOH, this is a good counterpoint by BLVCKLIGHTai of the same event, well worth reading from the perspective of the creators interested in this area.
Warp-as-History — Research project, but with code now, using video warping as a generation control mechanism. Kind of a world-modelesque thing, where they control the camera motion with arrow keys. Video demos, cool.
A couple links to new “world model” items: minWM: Minimal Interactive World Models — A minimal framework and tutorial for building real-time interactive world models, which you can do too!
minWM is our contribution to the world-model community: a full-stack open-source framework that walks you end-to-end through turning a bidirectional T2V foundation model into an action-conditioned video world model — with example data, runnable scripts, Claude skills capturing our hands-on experience, and onboarding knowledge for newcomers. We hope more researchers and developers join us in growing the community together.
Starchild-1: Real-Time Multimodal World Model — Odyssey's world model doing audio and visual generation in real time, research model/report.
Web fun (sorry Matt, running low this time): mykonos-island-voxels — Browser-based isometric island builder with the look of Mykonos. The demo isn’t live anymore, but the image is chef’s-kiss. Basically a thematically tuned version of a few of the others I’ve posted here in the past months.
SplatStroll Avatar — Walk Through Your 3D Gaussian Splatting Worlds — Add a walking avatar to Gaussian splat scenes. Interesting: the demo works with a premade avatar and several scenes, so you can see what it’s like to click on a bit of the landscape and have them walk there. It’s more compelling than I expected to drive them, but I am a gamer…
Tools
three-fluid-fx — Fluid dynamics and particle effect examples for Three.js.
StemStudio — open-source 3D editor and runtime for the web — “StemStudio is an open-source 3D editor and runtime built on Three.js. Behaviors, ECS lambdas, physics, AI copilot, and multiplayer — all local-first, with a UI you can ship.”
Sorceress 3D Studio — AI 3D Character Generator with Automatic Rigging — Generate 3D characters and get automatic rigging.
Current Rothko — A Rothko color field based on current weather. It sure wasn’t much varied during my heat wave last week. Via various people!
Spotify and UMG: Licensing for Fan-Made Covers and Remixes — Licensing agreements for fan creativity. Wasn’t sure where to put this, but here since I also wanted to put this in:
Trevor Paglen and Holly Herndon on Making Art with AI and What the Discourse Is Missing (via Garbage Day). Interviewed by Max Read for some reason, there are some interesting positions in this. Their discussion of “protocol art” sounds like procgen (procedural generation) rule sets: “When I [HH] say protocol art, what we’re talking about is that the site of art making is moving upstream of the media generation or the object creation, to a kind of logic or rule set or protocol that then allows the other things to come to fruition.”
And this concern about not having control of the actual knobs of the raw model is one I often feel too:
A lot of those programs have preprompting under the hood that you’re not seeing. Midjourney has all kinds of tags that make it look a specific way, a kind of beautiful way. Suno has tags you can’t see that make the output always sound “good.” You can’t interact directly with the weights of the model. You’re being confined so the output always sounds quote-unquote “good.” When Trevor and I are working with models, you can get way more gnarly because you have way more granular control. It’s just not being precleaned.
677 Novels Worth Arguing About (via Matt Muir). An infovis network display response to the Guardian’s silly top 100 books that were largely English canon, aren’t we done with that? Oh, right, it was 172 authors, academics, critics, and journalists who did the voting, not readers. Anyway, you can dive in here.
Microcosm Industries (also via Matt Muir) — Simulation toys/games, some of which I’ve had in past issues. Sim City, Townscaper, Tiny Glade, Factorio, etc.
👀 The NSFW History of Strip Mahjong Games (via Jacob Garbe) — This is pretty astounding. Absolutely nsfw including cartoony naked art. It’s full of amazing paragraphs like:
Not that it bothered Nihon Bussan too much, who managed yet another title in 1988 with Mahjong Satsujin Jiken, the first murder mystery strip mahjong game. By all accounts a standard entry to the genre but with a light detective story layered on top, the game of crime and sex was also the first from the company to showcase the effects of JAMMA’s pressure, characters mysteriously missing nipples, their naked bodies as smooth as a doll’s.
Vya — Thesis, Showcase 2026 (via Kottke) — An NYU game thesis project that uses a collage style. “Vya retells Dante’s Inferno as the story of a woman’s descent through the nine levels of midlife hell. Along the way, she sheds the skins that others have chosen for her, leaving her own skin free to undergo a transformative process of self-discovery.” There are other student games linked in the app showcase, too.
Lost Wiki: Kozlovka (on Steam; via Jonathan Cheng) — Explore a Wikipedia-like database and solve a small town mystery in Eastern Europe. You're a journalist in the 1990s scrolling through articles uncovering secrets. Neurocracy inspired, I assume? Overwhelmingly Positive reviews right now.
AI: Multi-Agent Scripts for Murder Mystery Games — Paper on using collaborative multi-agent systems to generate scripts for imperfect-information reasoning games. In other words, some of the agents don’t know everything, some are lying, etc. I guess you can work on any problem with enough agents, now…
The AI fiction prize winning cluster-f*ck: AI is Writing Prize-Winning Fiction (Pangram Labs) • ‘Obvious markers of AI’: doubts raised over Commonwealth Short Story Prize winner (Guardian) • ‘My writing process is unusual,’ says prize-winning author accused of being AI (Observer) • AI-slop, GrantaGate, and Bad Writing (Tuhin Chakrabarty) — All recent articles saved on the same Granta prize winner that fails various AI writing checks and tells. (There are more, I stopped at a point. I felt I had to cover this, but…)
Tuhin Chakrabarty and a smart student run it through Infinigram (demo here) to find that fanfic in pretraining is a major signal for these tells. But also generic internet, cuz, of course.
I am not a fan of AI detectors, as I think I’ve said. I like semicolons and emdashes, too. And I find it tiring that people are so overjoyed to be outraged by the prize givers not detecting this themselves. (I did not read it. I’ve got things to do here, like finish Planet Crafter, see art…)
😨 Hating on fanfic style aside, I’m extremely disturbed by the crackdown on writers of slash fanfic in China and now Russia too: “Russian woman sentenced to labour over gay Stray Kids fanfiction.”
👍 Narrative Nubs v1.0 — Nick Montfort’s new release of “nubs” of famous narrative procgen systems:
Narrative Nubs is a project to develop a set of story generation systems. These nubs (as in “the nub of the argument”) are:
Based on existing (including historical and present-day) story generators, e.g., TALE-SPIN, TAILOR, MINSTREL, DAYDREAMER, Story Machine, Grandmother, and MEXICA. These systems are just examples; many factors will determine the specific nubs implemented and the order in which we progress.
Abstractions, or miniatures as they were called in Roger Schank’s lab at Yale, which are significantly more concise than the original systems and capture the main interesting things about them. Given this, they don’t work in exactly the same way.
Systematic. They are all coded in the same language (Python 3) in a consistent style, and share libraries when suitable. E.g., there is a single realizer implemented as a library to serve all the nubs that generate sentences using templates.
NuExtract3, a 4B vision language model, is good at structured extraction (json) from documents, reports Daniel van Strien. I’ve posted his book in progress, AI Patterns for GLAM, right? (Galleries, libraries, archives, museums.)
The April 2026 Web by the Numbers — Query 2.19 billion Common Crawl pages with DuckDB, zero download required. Via Danien van Strien.
From Aaron Cope, museum vector embedding releases: Shared vector embeddings updates. There is a lot of ground to cover in this blog post: The Met publishing their own vector embeddings, SFO Museum publishing 1152-dimension vector embeddings for its images, SFO Museum producing 1152-dimension vector embeddings for NGA and MoMA, and a whole bunch of updates to the tooling used to generate and query vector embeddings targeting local-first and consumer-grade hardware.
Violin - Together.ai’s open source video translation skill, using a few cheap models. It’s clever and useful.
Full Screen Dashboard Template Chart Transitions by Lao — Animated full-screen dashboard template with chart transitions buttery smooth, on Observable. Really nice.
Agent skills/tools:
tufte-viz Claude Code skill — A Claude Code skill applying Edward Tufte data visualization principles.
MMSkills Skill Library — A library of multimodal AI skills for agents.
Text within this block will maintain its original spacing when published
Our inner lives gathering, gathering while angels disguise themselves as a kitchen sponge. Dear angel, I mutter! I address the rushing world. I make saints of soda cans and fly north in winter. Every person is an address the way every person is a calendar. You are a Wednesday afternoon on Rue Etienne Marcel. On the last day of May, a soccer game is won and fireworks burst in the streets. Young pigeons, in fright, flee a stone church in Belleville.
— From “Every Person Is an Address, Every Person Is a Calendar” by Megan Fernandes.
Happy June! How is it June?!
Best, Lynn (@arnicas on mostly bluesky, mastodon, ex twitter).
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.