§ A · The one principle
the test every tool has to passEvery tool makes one invisible mechanism something you can see, change, and break.
Not "teach about AI" in general — take one hidden thing (a probability distribution, a denoising step, drift across frames) and turn it into a knob you can move. A tool that explains is a slideshow; a tool that lets you turn one knob and watch the output move is an instrument. The camp builds instruments. The companion rule: the tools are simulations on purpose — frozen examples and hand-authored distributions, not a live model — so the mechanism stays legible and the demo is identical every time. A live model would be more impressive and less teachable.
§ A.1 · The shape of it
what "an instrument" cost, in parts§ A.2 · How AI was used
development · documentation · source materialLearning Machines was developed with AI assistance for planning, coding, copy drafting, research, and critique. AI-assisted workflows also helped organize session notes, transcripts, chat exports, and aggregate program data. Public recaps, quotations, attributions, and findings are reviewed by the facilitator against source records. Most interactive instruments on this site are authored browser simulations, not live machine-learning models. Genuine model outputs are labeled with the source information that was preserved; incomplete provenance is named rather than guessed. External services process submitted material under their own terms and account settings.
What is not AI
Most Learning Machines instruments do not run machine learning. They use authored data, deterministic JavaScript, rules, illustrations, and human input.
Some parts of the site use ordinary web services: Google Forms collects submissions, GoatCounter provides basic traffic counts, and a few activities retrieve Wikipedia content. Notes and preferences may also be stored locally in the browser. None of these features generates content with AI; each has its own data and privacy considerations.
Model-output provenance
“Pre-generated” is not enough to identify an example. Genuine model outputs should record the product and model, date, exact prompt, relevant settings, whether the example is complete or excerpted, whether the facilitator edited it, and whether it contains participant material.
Visible source labels used across the project are: Authored teaching simulation; Frozen real-model output · model/date documented; Edited or excerpted model output; AI-assisted text · human-reviewed; Participant-created artifact; and Live external model. When a historical record is missing, the label is Provenance incomplete · do not attribute to a named model rather than a guessed attribution.
§ B · The constraints & what each one buys
every rule is a pedagogical decision in disguiseSeven rules shaped every tool. None is about taste — each one buys something specific: it keeps the demo reliable live, keeps the tool equitable to reach, keeps the mechanism legible, or keeps the file alive for years. The column on the right names what each constraint is really protecting.
What the rule-set optimizes for, counted
Tally the right-hand column and the priorities show themselves — this is a curriculum that would rather be dependable and reachable than clever.
§ C · Where they came from
two bodies of work, joinedA generative-AI course
ELIZA-vs-LLM comparisons, next-token prediction, image Default Tests, diffusion and latent-space activities, Temporal Telephone, A/B/C prompt testing. The activities that survived contact with real students were the ones strong enough to adapt for adults — which is why the camp's tools feel classroom-tested: they are.
CC Fest Coding Camp
Free, virtual, community-centered workshops — beginner-friendly but intellectually serious, with recaps, asynchronous access, guest speakers, assignment tiers, and a closing showcase. Learning Machines is the two lineages fused: classroom-tested mechanisms, delivered the CC Fest way. The full account is in the Project Brief.
§ D · The method inside each tool
the loop the tools are shaped to supportThe take-home isn't any one tool — it's the loop. Predict, change one variable, compare, name precisely what the machine did. Each session widens the same four beats into a full investigation. Participants forget the URLs; they keep the method.
§ E · Make your own
the "Explain" studio pathwayBecause the tools use inspectable local source rather than a bundled application, a direct way to understand one is to open its entry point and follow or change it — swap the frozen examples for your own, relabel a knob, or fork the structure into a new mechanism. Some tools are one file; others draw on shared assets or additional local modules. Building an explanatory tool, poster, or concept bridge is one of the camp's final-project pathways.