AI Literacy: Learn to Question, Test, and Evaluate AI
AI is changing how people research, create, communicate, and work. Using it effectively involves more than writing prompts: you need to check its results.
This free introductory class teaches one practical skill: writing clear criteria to evaluate AI-generated responses. You do not need to become an AI engineer to understand what AI does or recognize its limitations.
No programming experience is required. Explore without an account or job application. Work at your own pace on a phone, tablet or computer; a paid subscription is unnecessary.
What you will learn
- Explain a model in everyday language and distinguish prompts, input files, responses and rubrics.
- Identify limitations and misleading AI claims.
- Write atomic, verifiable, objective and self-contained PASS/FAIL criteria.
- Choose positive and negative weights and distinguish explicit from implicit requirements.
- Score fictional work using evidence and explain one improvement.
- Understand how evaluation relates to different kinds of AI-assisted work.
What does a model do?
A model is software trained to find patterns in examples. A text-generating model uses patterns and the context you supply to produce a response, one piece of text at a time. It can draft, summarize and transform information. A fluent response can still contain invented details or mistakes.
- Prompt
- The instruction or question you give the model.
- Input file
- Material supplied with a task, such as a fictional price list.
- Model response
- The generated answer or deliverable you inspect.
- Rubric
- A set of criteria: clear rules for deciding whether the response meets requirements.
- Evaluation
- Checking output against those rules and recording evidence.
- Hallucination
- A generated claim that sounds plausible but is unsupported or false.
Start by checking names, numbers and citations against the task facts. OpenAI’s evaluation guidance explains why repeatable tests and human judgment matter.
What people hear about AI — and what the evidence actually shows
“AI knows everything.”
Models have limits in training, context and access to current sources. A tool connection can help retrieve information, but still needs checking.
Try it: Ask for a fictional vendor’s opening hours: the answer must use the supplied schedule, not invent one. Stanford AI Index: uneven capabilities.
“AI always tells the truth.”
A model can produce unsupported statements. It does not guarantee that each sentence was fact-checked.
Try it: A convincing citation may not exist; open it and check the claim. NIST AI risk management resources.
“You need to know how to code to work with AI.”
This class uses careful reading and writing. Certain evaluation or quality-review tasks use research or domain knowledge; employer and project requirements vary.
Try it: You can compare a fictional itinerary with closure dates without writing software. Educational evaluation examples.
“Everyone working in AI is an engineer.”
AI-assisted work includes writing, research, education and quality review alongside specialized technical careers. Some roles require advanced credentials.
Try it: A teacher can review whether a generated lesson meets a learning objective. Teacher skills and rubrics.
“AI is replacing every job.”
Tasks and occupations differ. BLS projects growth in some occupations and declines in others; a projection is not a guarantee.
Try it: Software developers and customer service representatives have different projected trends. BLS occupational projections.
“AI will automatically create new opportunities for everyone.”
Access, training and job changes are uneven. Stanford reports early-career employment declines in some AI-exposed work, with limitations in the data.
Try it: A growing occupation can still be difficult for a new entrant to access. Stanford employment evidence.
“You need expensive equipment or a paid subscription to learn AI.”
You can complete this lesson with the fictional response on the page, without calling any AI service. Other courses or tools may have different requirements.
Try it: Grade the supplied response on your current device, or write your criteria on paper. Free practice in this class.
“An AI-generated answer is correct because it sounds convincing.”
Writing style is not evidence of accuracy. Use specific tests and inspect supporting facts.
Try it: A polished itinerary still fails if it schedules a visit on a closure day. Evaluation best practices.
AI and Work: What the Data Tells Us
United States · projected employment change, 2024–2034. These percentages compare expected employment in 2034 with employment in 2024. They are not counts of current AI vacancies, and these occupations include work beyond AI.
| Occupation or group | Projected growth |
|---|---|
| Computer and mathematical occupations | 10.1% |
| All occupations | 3.1% |
| Data scientists | 33.5% |
| Software developers | 15.8% |
Sources: BLS overview (occupational group) and BLS selected occupations (other rows).
Static reviewed snapshot, verified 7 November 2026. Check source dates before relying on it.
Population: U.S. wage and salary employment across industries plus self-employment. BLS published the overview in January 2026 and the selected-occupations article on 16 July 2026. The snapshot is periodically reviewed, not live or real-time.
BLS also projects declines for some administrative and customer-service occupations. Its long-term projections do not predict an individual’s job outcome.
Contrasting evidence: Stanford’s 21 October 2026 report describes falling employment among young workers in highly AI-exposed software-development and customer-service work. The underlying payroll data do not represent every worker, and association with AI exposure alone does not prove AI caused a change. Monthly observations and decade-long projections answer different questions.
Different kinds of change: automation hands a task to software; augmentation helps a person do it. Existing occupations may acquire checking responsibilities. Specialized AI careers can require advanced study. Entry-level evaluation tasks may emphasize careful reading, writing or subject knowledge; availability and requirements vary.
For further context, explore the Stanford AI Index and Lightcast’s job-posting analysis. Postings measure advertised demand, not completed hires or guaranteed openings. These sources do not endorse this class.
Download the versioned data and source metadata. If a source becomes unavailable, this dated snapshot remains visible; missing values are never estimated.
A small evaluation reference library
Links and descriptions checked 7 November 2026. Start with this class before the technical references. External organizations are reference providers, not partners.
- Cornell: Using rubrics — Beginner. A teaching guide to measurable criteria, feedback and revision; not an enrolled class.
- OpenAI Evaluation Best Practices — Intermediate. Concepts for designing tests; API examples are optional. This class does not use its hosted Evals platform.
- Microsoft Evaluation Guide — Intermediate. Planning and interpreting agent evaluations.
- Anthropic educational evaluation examples — Beginner to intermediate. Read examples of educational rubrics; running repository tooling is optional.
- UK AI Security Institute Inspect scoring — Advanced. Software scoring framework; Python knowledge helps.
- Google ADK evaluation criteria — Advanced. Agent evaluation metrics and technical configuration.
Where you could apply this skill
Optional examples: check a PIXIE handoff for stated next actions; evaluate a synthetic educational exercise for 50 Ways to Leave Another; verify a Made-Sick resource description against its cited source; or check Duet and game documentation against specified rules. These are contexts for practice, not new integrations or assignments.
Use the existing portfolio-evidence workflow to explain your own contribution. GitHub remains the shared record; no second tracker is needed.
Interested in Exploring AI-Related Work?
Some organizations offer paid projects involving AI evaluation, research, writing, and specialist knowledge. Opportunities vary by location, experience, availability, and project requirements.
Handshake is one place to explore potential opportunities. Applying is optional, and completing this class does not guarantee acceptance or paid work.
Explore Handshake (referral link)
Referral disclosure: This is a referral link. Loptr Lab's founder may receive compensation if an eligible referral completes qualifying paid work. You are not required to use this link or apply to participate in the training.
Before applying, check Handshake’s current referral conditions and the requirements of the specific project. Referral rewards are separate from learner compensation. Loptr Lab does not represent itself as an authorized Handshake training provider or recruiting partner.
Independent, non-referral route: research occupations through the BLS Occupational Outlook Handbook and check requirements directly with employers.