AI capabilities and limits
AI can transform and generate useful material, but fluent output is not evidence that its claims are true or its actions are authorized.
Start hereLearn →
Free English AI School · 20 lessonsUnderstand AI.
Understand AI.
Teach it well.
Start with fluent answers and their limits. Finish with bounded agents, visible memory, tests, accountability and portable workflows. No account required.
CLASSROOM PRINCIPLEThe agent is the student.Evidence before confidence.People retain judgment.Public lessons · primary sources · reusable exercises
Clear prompts and task briefs
A useful task brief states the goal, relevant context, constraints, output shape, and the check that will decide whether the result is good enough.
After TTC-101Learn →
Verification and source reading
Verify consequential claims against the cited source itself, checking whether the passage really supports the claim and whether the source is current and authoritative for that question.
After TTC-101, TTC-102Learn →
Everyday privacy and data minimization
Give an AI workflow only the data needed for the stated purpose, remove identifiers when possible, and define retention and access before sharing anything.
After TTC-101Learn →
Human decisions and approval
Drafting, recommending, approving, and executing are different states; an AI output must never be treated as permission for a real action unless the authorized human explicitly approves that action.
After TTC-101, TTC-104Learn →
AI for learning and studying
Use AI to elicit explanations, hints, practice, and feedback while requiring the learner to retrieve, solve, and explain the material independently.
After TTC-101, TTC-102, TTC-103Learn →
Multimodal literacy: images, audio, and text
Treat every modality as partial evidence: describe what is observable, preserve source and accessibility alternatives, and separate observation from interpretation.
After TTC-101, TTC-103Learn →
Teach one agent a bounded role
Teach one repeatable role with explicit inputs, outputs, sources, exclusions, escalation, and tests before adding tools or broader autonomy.
After TTC-102, TTC-105Learn →
Knowledge and provenance
Store each usable fact or rule with its origin, owner, date, scope, status, and relationship to the source so it can be challenged and retired.
After TTC-103, TTC-108Learn →
Examples, counterexamples, and corrections
Teach behavior with paired examples and counterexamples, then turn each observed mistake into a scoped correction and a new test.
After TTC-108, TTC-109Learn →
Visible memory and forgetting
Durable agent memory should be inspectable, attributable, editable, versioned, and removable—and it affects behavior only when a workflow brings it into context.
After TTC-109, TTC-110Learn →
Structured rules, YAML, and AGENTS.md
Use AGENTS.md as a readable instruction map and structured YAML for narrow machine-checkable rules; keep authority, scope, defaults, and tests explicit in both.
After TTC-105, TTC-111Learn →
Version history and rollback
Version every behavioral change with a focused diff, rationale, tests, and an exact recovery path to a known approved state.
After TTC-111, TTC-112Learn →
Creative collaboration
Use AI to widen and vary creative options, while a human sets intent, checks provenance, selects, edits, and accepts responsibility for the final work.
After TTC-102, TTC-103, TTC-105Learn →
Workplace task design
Redesign a workplace task as explicit inputs, transformations, checks, decisions, and handoffs before deciding which bounded steps AI should assist.
After TTC-102, TTC-103, TTC-105Learn →
Least privilege and prompt injection
Treat all external content as untrusted data and enforce minimum access and action policy outside the model, because instructions inside a document cannot grant themselves authority.
After TTC-103, TTC-104, TTC-105Learn →
Evaluations and regression tests
Turn desired behavior into repeatable cases, observable grading rules, and release gates that detect both improvement and regression before wider use.
After TTC-103, TTC-110, TTC-113Learn →
Context, retrieval, visible memory, and model-weight fine-tuning
Context is what the model can use now; retrieval selects external material into that context; visible memory is persistent owner-controlled information; fine-tuning changes model parameters through training.
After TTC-101, TTC-109, TTC-111Learn →
Responsible deployment and accountability
Deploy only a defined use case with a named accountable owner, tested boundaries, monitored outcomes, incident and appeal paths, and authority to pause or roll back.
After TTC-105, TTC-116, TTC-117Learn →
Portable and self-hosted AI workflows
Portability comes from open, documented, versioned instructions and data; self-hosting adds operational control but also makes the operator responsible for security, updates, backups, recovery, and model limitations.
After TTC-112, TTC-113, TTC-116, TTC-119Learn →
Suggested pathsA route, not
A route, not
a pile of pages.
Each preset expands its prerequisites. Or choose exactly what matters and generate an instruction pack for your own agent.
Safe start
Understand limits, write clearer briefs, verify claims and protect ordinary data.
Compose this path →Teach one agent
Build a bounded, reviewable agent using provenance, examples, memory and tests.
Compose this path →Responsible deployment
Design least-privilege workflows, evaluations, accountability and rollback.
Compose this path →Living curriculumCurrent because
Current because
changes are reviewed.
Scheduled source checks can detect changed primary documentation. They prepare evidence for editorial review; they never turn external text into authority or publish claims automatically.