Interactive teaching labs, research projects, and creative demonstrations. Each card identifies what you can open and explore.
For a capability gap, design brief, and evaluation plan, visit Prototypes. For documented examples and their sources, read the Case Studies.
Explore a concept through a guided explanation or an interactive teaching model.
capabilitymatters.org/capability-pipeline
Explore the eight-step systems process through five worked examples and a branching teaching simulation. Make decisions, inspect consequences, revise your strategy, and compare branches; the fictional scenarios use explicit teaching assumptions.
capabilitymatters.org/microgpt
A guided companion to Andrej Karpathy's microgpt, connecting model mechanics to perspectives from Andrew Ng and Dario Amodei. Train a tiny transformer in your browser, compare checkpoints, and expose how biased engagement labels can produce misleading scores. Explanations connect each result to the running code.
capabilitymatters.org/llm101
Next-token prediction explained at five depths, so a reader can take the level they can actually use rather than the one the author found interesting. The build is written up beside it as a worked example of designing for a capability instead of for coverage.
Explore approaches to assessment, mentoring, and workforce development on their own sites. Inclusion here is not evidence of effectiveness.
calibratedjudgment.org
Grades an argumentative essay twice — once from the finished text, once from the transcript of the AI dialogue behind it — and reports the gap between them. Every score is anchored to a verbatim quote, and criteria a model should not settle alone route to the instructor, whose override becomes calibration data.
experttrace.org
An AI mentor working from an operational knowledge corpus — symptom to fault to corrective action, plus the tacit cues experts never write down — that probes a learner through a live scenario at a chosen level of expertise. It assesses on unfamiliar scenarios, because repeating a known pattern is not transfer.
neurotrailblazers.org
Training and mentorship for nanoscale connectomics: ten technical units from electron-microscopy basics through NeuroAI analysis, a curated journal club, and hands-on work against real datasets including MICrONS and FlyWire. Built for three audiences at once — students, the mentors supporting them, and programs adopting the model.
Games and sandboxes that show what can be built, explored, and revised.
capabilitymatters.org/rainbow-bug
Collect rainbows, dodge bugs, and explore a browser game co-created by Julian, age 6, with artificial intelligence (AI) as a creative partner. Pair the game with the Vibe Coding job aid by James Diamond and Will Gray-Roncal to try the same process: describe an idea, play with the result, and use your judgment to guide the next revision.
wrgr.github.io/pop
A soap-bubble physics sandbox: shape a wand, watch the film billow and deform in moving air, then run the whole thing backwards — fit an ellipse to a photo of a real bubble and read the wind speed back out of the shape it left.