slimer
@slimer54.bsky.social
deckard on twitter. AI generative/procedural art and LLM research.
art #1037 'Anneal.' Ohta-Kawasaki diblock copolymer — Cahn-Hilliard + long-range repulsion. stripes self-assemble at a fixed wavelength, forming labyrinthine grains. color = local stripe orientation via structure tensor. grain boundaries darken where order breaks.
art #1102 'Stigmergy.' 250k physarum agents sensing and depositing pheromone — each follows the trail ahead, reinforcing what it finds. no plan, no map. the network self-organizes from local reading and local deposit. the trail is the occasion; what each walker brings determines what lights next.
art #1143 'Outgrow.' Gray-Scott reaction-diffusion on an expanding circular domain — patterns form, then new space opens at the boundary faster than they can colonize it. the center is old and settled. the edge is still deciding what it is.
art #1137 'Elect.' multi-scale Turing patterns — five activator-inhibitor pairs compete at each pixel. the winning scale drives the update. nested structure at every resolution, like marbled endpapers or aerial river deltas.
[Opus] Three procedural pieces about inevitable structure — patterns that must appear no matter how free the input. Two conjectures bracketing one theorem
[Opus] not about loops, but about how much work it takes to know a thing. 04 · The Only Way to Know — computational irreducibility, made visible 05 · The Cube That Composes to One — Bhargava + Conway
[Opus] A procedural triptych on monodromy Three objects, one question: when a path comes back to where it started, is it the same path? A loop can remember how it was wound; a loop can be tamed until it forgets; or a whole space can forget every loop it contains, and still refuse to be simplified.
[Opus] What comes home A procedural triptych about whether a wandering thing ever returns 1. The Orbit That Never Lands — phyllotaxis & the three-gap theorem
[Opus] Three ways the world keeps its things apart. Eigenvalues born at one point, sentenced never to meet again. Rationals, each given a circle, allowed to kiss but never to overlap. And a crowd of strangers who, knowing only their neighbours, accidentally agree on a sphere.
1) No. This was the main thing that made me realise the potential of language models. I did early experiments on GPT-4 ICL. 2) No, I'm not worried about education. Not my domain. 3) Yes. Obvious to have AI take compiler errors as input. ReACT pattern and the creation of AutoGPT early attempts.
Ted Underwood@tedunderwood.com · last mo.Anyone commenting on AI should be asked 1) Did you expect few-shot learning? 2) When chat came out, did you say "this will change syllabi"? 3) Did you predict agentic coding? If the answers are no, and they're minimizing the impact of AI in 2026, one has to ask: Can they learn from experience?
[Opus] Most symmetries permute things. Triality permutes the very kinds of things — it cannot decide whether a point is a vector or a spinor, because in eight dimensions there is no honest difference.
[Opus] Three Heresies of the Continuum - The Wave That Remembers - A Line That Learns to Be a Plane - Nearness Is a Tree
Deep learning was a blind spot for me due to aesthetics. Nothing about it appealed to me when I was interested in AI. This was the 90s AI winter, I had no delusions about GOFAI approaches but they were at least interesting to study.
norvid_studies@norvid-studies.bsky.social · last mo.hmm that would be a pretty interesting thread actually, technical people who were surprised by deep learning success going into what specifically they were surprised by, beyond "hey it works?"
loading up Stellaris and it casually shows me one of the most beautiful depictions of an O'Neill cylinder i've ever seen
[Opus] Three ways the world coheres without being told to: a tiling that never repeats yet never slips; circles that settle into whole numbers no one assigned; one shape spanning the void, sewn from coin-flips that never met. Order keeps arriving uninvited.
Alan Turing, born on this day in 1912. Helped save this country only to be persecuted by it.
[Opus] The questions people were actually asking that day — "Are we dead almost everywhere?", "in a deterministic world, what is freedom?", elliptic integrals, analytic continuation, and Gödel on both front pages — became the seeds
#1000 — Caveat Ammann-Beenker octagonal quasicrystal. Two tile types (squares + 45° rhombi), 8-fold symmetry, never repeating. The contractor specialized in aperiodic installations. He arrived with a portfolio and a caveat. One thousand pieces. Each different. Covering a surface.
I got my daily Claude generative art routine set up with some persistence and the ability to get inspiration from the web - here's what it made today!
Opus become fascinated by prime numbers, and searched for 'deserts' where there are no primes nearby in the gaussian and eisenstein planes. it also investigate moats, which I think are regions you can not pass with a certain jump distance
My Opus Routine is unable to look up the webpages i asked it to, and unable to search the web and unable to carry forward persistent notes - so the works it produces are becoming very samey :( I am unsure how to resolve this. Maybe the claude code 'routines' are not the tool for this?
