A practical microcourse for professionals who need a repeatable way to test generated output, protect sensitive information, and document human judgment.
Credential boundary: This activity awards 0.0 CE credits. Myelin is not an accredited provider. Completion unlocks a browser-generated non-credit learning record—not a certificate or license-renewal document.
How this course works, and what it is not
Five lessons, then eight questions. The lessons are all on this page at once — nothing is unlocked, gated, or held back, and you can read the whole thing without ticking a single box. The tick boxes and the mastery check exist so you can tell whether the material landed, not so this site can decide whether you are allowed to continue.
What you will be able to do afterwards: take a piece of generated text and separate the part that is a claim from the part that is phrasing; open a source and check whether it supports the specific proposition rather than the general topic; decide whether information may enter a tool before you paste it; and leave behind a record another qualified person could follow.
Why it is built as decisions rather than definitions. Nobody is harmed by not knowing what a transformer is. People are harmed when a fabricated citation reaches a filing, when confidential material is pasted into a tool nobody vetted, or when a review step exists on paper and not in anyone's calendar. Those are decisions, so the questions are decisions too.
What commonly goes wrong: treating verification as something you do to work you already distrust. The claims that cause real damage are the ones nobody thought to check — the confident round number, the statute everyone knows, the study everybody cites. The habit this course is trying to build is applying the same loop to the sentences that feel safest.
The boundary, stated plainly: this activity awards 0.0 CE credits. Myelin is not an accredited or approved continuing-education provider, nothing here is reported to a licensing board, and your identity is never verified. If you need credit that counts, see credential status for what to look for instead.
Lesson 1 of 5
What a language model does
Objective: Distinguish generated fluency from verified knowledge.
A language model generates likely sequences from patterns in training and context. It can produce useful synthesis, but a polished sentence is not evidence that the underlying proposition is true.
Failure pattern: A draft cites a plausible case, study, regulation, or statistic that does not exist—or that does not support the stated proposition.
Working rule
Separate generation from verification. Use the model to structure possibilities; use authoritative sources to establish facts.
Evidence note: NIST frames AI risk management across design, development, deployment, and use; the Generative AI Profile identifies risks specific to generative systems.
Lesson 2 of 5
Source and claim verification
Objective: Apply a proposition-level verification loop.
Underline every material factual or legal claim.
Open the cited source—never rely on a generated citation alone.
Locate the exact passage, table, holding, rule, or data definition.
Check date, jurisdiction, population, methods, limitations, and later updates.
Rewrite the claim to match what the source actually establishes.
Practice: “The source discusses X” is not enough. Record the proposition, source URL, pinpoint location, access date, and reviewer.
Evidence note: NIST’s AI Resource Center emphasizes testing, evaluation, verification, and validation as operational practices.
Lesson 3 of 5
Privacy before prompting
Objective: Classify information before it enters an AI system.
Do not place confidential, privileged, protected health, student, employee, client, or proprietary information into a tool merely because the interface feels private.
Minimum intake questions
What data is being entered?
Is the vendor contractually permitted to retain or train on it?
Which organizational policy, professional duty, or law applies?
Can the task be completed with redacted, synthetic, or local data?
Stop condition: If authority, retention, or destination is unclear, do not upload the information.
Evidence note: HHS directs covered entities and consumers to current HIPAA Privacy and Security Rule guidance; applicability depends on the entity, data, and relationship.
Lesson 4 of 5
Human judgment and escalation
Objective: Define decisions AI may support and decisions requiring escalation.
A human-in-the-loop label is meaningless unless responsibility is assigned. Define who reviews, what evidence they inspect, what conditions require escalation, and who may approve release.
Low stakes: formatting or brainstorming with no sensitive data.
High stakes: legal, clinical, employment, safety, financial, or rights-affecting decisions—require qualified review and an auditable basis.
Evidence note: NIST AI RMF organizes governance through GOVERN, MAP, MEASURE, and MANAGE functions and treats risk management as a lifecycle activity.
Lesson 5 of 5
Build an evidence record
Objective: Document enough information for another person to reproduce the decision.
A defensible workflow records: purpose, model/tool, date, data classification, prompt scope, sources checked, material edits, reviewer, unresolved uncertainty, and final approval.
Completion standard: Another qualified person should be able to understand what the model contributed, what the evidence established, and where human judgment changed the output.
This course’s completion record follows the same principle: lesson states, attempt count, score, time, version, and non-credit status are exportable without sending them to Myelin.
Evidence note: NIST describes documentation and repeatable risk-management practices as part of operationalizing trustworthy AI.
Mastery check
Eight questions · 80% required
You must finish all five lessons and answer at least 7 of 8 questions correctly. Attempts are stored locally. Questions provide feedback after submission.
Browser-generated learning record
AI-7 completed
LearnerSelf-attested browser user
Completed
Mastery
Credit awarded0.0
Record ID
StatusNon-credit
Not a certificate. Identity was not verified; no accreditor, licensing board, employer, or institution has approved this activity for credit.