The words we use, defined plainly. Basic to advanced, from what a prompt is to the integrator that keeps PHYS//LAB's orbits from falling apart. New terms get added as the magazine grows.
Art created wholly or partly through an autonomous system - an algorithm, a set of rules, or a trained model - that the artist sets in motion rather than directly controls stroke by stroke. It predates AI by decades: plotter art, cellular automata and generative music all came first. AI image and text models are its newest, most visible branch, not the whole field.
Visual art made with machine learning models - most often text-to-image diffusion models - as the generating tool. The term covers everything from a single prompt typed into a public tool to elaborate multi-stage pipelines mixing several models with manual editing on top.
The text, and sometimes image, instruction fed to a generative model to describe what it should produce. Wording, order and specificity all measurably change the output, which is what makes prompt writing its own skill rather than just asking nicely.
The starting number a generative model's random number generator uses. The same prompt with the same seed, on the same model, produces the same result; change only the seed with everything else held equal, and the composition reshuffles while often keeping a family resemblance to the original.
The dominant architecture behind current AI image generators - Midjourney, Stable Diffusion, DALL-E among them. Trained to reverse a process of gradually adding noise to an image, so at generation time it starts from pure noise and denoises it step by step into a picture that matches the prompt.
The compressed, abstract numerical representation a model works in internally, rather than raw pixels. Two images that are conceptually similar - two portraits, two sunsets - tend to sit close together in this space, which is what lets a model interpolate smoothly between them instead of jumping abruptly.
The practice of deliberately structuring, ordering and wording prompts to reliably steer a generative model's output. It exists as its own skill because the same underlying request can produce wildly different results depending on phrasing alone.
A term coined for images made by writing prompts rather than by drawing or photographing - treating prompt-based image generation as its own medium, with its own craft, rather than a shortcut version of illustration or photography.
Applying the visual character of one image or artist's body of work - palette, brushwork, composition tendencies - onto different subject matter. Modern AI tools usually do this with a reference image or a trained style model, rather than manual imitation.
A creative tradition built on openly reworking existing material - sampling, mashups, fan edits, and now AI-assisted reinterpretation - that treats culture as something to build on rather than only originate from scratch. "Everything is a remix" is our own working premise, not just a slogan we liked the sound of.
Workflows where a person and a generative model each do the part they're better at - a human sets intent, taste and final judgment; the model handles combinatorial exploration a person couldn't do by hand at the same speed - rather than either working alone.
Turning data into sound so it can be heard instead of read - the audio counterpart of a chart. The mapping is the entire design decision: which measurement drives which musical parameter, and how far it is allowed to push it. LUCKY is a worked example, putting wind speed on tempo and precipitation on percussion.
Music produced by a system following rules rather than composed note by note and fixed in place. Each playback is a fresh run of the process, so no two are identical. Feed those rules live data, as LUCKY does with weather, and the piece cannot repeat even in principle: the inputs have moved on.
A simulation where every object exerts a force - usually gravity - on every other object simultaneously, rather than orbiting one fixed center. Cost grows with the square of the object count, which is exactly why PHYS//LAB's own scenarios cap out where they do.
Full control reference in the Docs →
A method for stepping a physics simulation forward in time that updates position using both velocity and acceleration, then splits the velocity update into two half-kicks around a fresh force calculation. It conserves energy far better over long runs than the naive alternative, Euler integration - which is why orbits in PHYS//LAB don't quietly spiral apart the longer you let them run.
How we actually built the integrator →
A small constant added before taking a square root in the inverse-square gravity law, so two bodies passing very close together don't divide by a number near zero and get flung to infinity in a single frame. It trades a little physical precision at extremely close range for a simulation that never explodes.
Creating content - textures, terrain, entire structures - algorithmically from a set of rules and some randomness, rather than by hand, so a small program can produce far more variation than anyone placed by hand ever could.
A way of drawing 3D space on a 2D screen with no vanishing point, so parallel lines stay parallel and nothing gets smaller with distance. PHYS//LAB renders this way by default. It's also the classic look of early strategy games, for the same practical reason - it's cheap to draw and easy to read at a glance.
An earlier approach to AI image generation, pre-dating diffusion models' current dominance, built from two competing networks - a generator making fakes, a discriminator trying to catch them - trained against each other until the fakes pass. Mostly superseded by diffusion models for general-purpose image generation, but still influential in specific tasks like fast style transfer.
The study and practice of building systems whose output gets judged creative by human standards - novel, valuable, unexpected - and of what that judgment even means once a machine, not just a person, is the one generating the candidates.
Complex, often unpredictable group-level patterns arising from simple local rules with no central controller - flocking, swarming, cell-like clustering. PHYS//LAB's CELLS scenario is a working example: each cell follows one simple turning rule, and clustering behavior emerges from thousands of them following it at once, not from any rule that actually says "cluster."
The philosophical view that consciousness, or something like it, is a fundamental and ubiquitous feature of reality - present in some minimal form even at the level of electrons and atoms - rather than something that only switches on once matter gets sufficiently complex, as in a brain.
David Chalmers' 1995 framing of the actual puzzle: even a complete physical account of the brain - every neuron, every signal - doesn't by itself explain why any of it is accompanied by subjective experience at all, rather than happening with no one home. Panpsychism, idealism and quantum-consciousness theories are all, in different ways, attempts to answer it.
In quantum mechanics, the genuine and well-established fact that certain measurements disturb the system being measured. Popular writing regularly stretches this into "consciousness collapses reality," but the effect is about physical interaction with a measuring instrument, not about a mind looking at something - worth keeping the two separate.
The real quantum mechanics behind PHYS//LAB →
A contested hypothesis, proposed by physicist Roger Penrose and anaesthesiologist Stuart Hameroff, that consciousness arises from quantum processes inside microtubules in brain cells. It has real defenders but sits well outside mainstream neuroscience and physics, both of which mostly consider the brain too warm and noisy for the quantum states involved to survive. Included here as a real, nameable position in this debate, not an endorsement of it.
The proposal, most associated with philosopher Nick Bostrom, that what we experience as physical reality could itself be a computation running on some deeper substrate - a simulation, in the same sense PHYS//LAB is a simulation, just one nobody outside it can inspect the source code of.
The philosophical position, older than quantum mechanics by centuries, that mind or consciousness is the fundamental stuff of reality and the physical world is somehow derived from or dependent on it - the mirror image of materialism, which puts matter first and treats mind as something that emerges from it. Panpsychism sits between the two: matter is still fundamental, but so, in some minimal way, is mind.
Read the field notes →