Go beyond ChatGPT. Understand how AI works, evaluate it critically, and build with it.
By the end, parents should be able to see clear evidence that students understand AI, evaluate it critically, and use it responsibly.
My child understands why AI works, not just how to use ChatGPT.
They can clearly explain core AI concepts like training, inference, tokens, context windows, hallucinations, RAG, AI agents, and multimodal AI.
They know how to evaluate new AI tools critically instead of chasing every trend.
They can recognize when AI is useful, when it may be unreliable, and when human judgment is still necessary.
They can use AI responsibly while considering privacy, bias, misinformation, and academic integrity.
Most students can open ChatGPT and type a prompt. We teach the technology behind them: how models learn, how they generate outputs, where their limitations come from, and how modern AI products are built.
Students are already surrounded by recommendation systems, chatbots, image generators, and AI study tools. This lab gives them the vocabulary and judgment to understand what they are using.
The goal is not to teach students to outsource thinking. Students learn to question AI outputs, verify claims, recognize bias, and decide when human judgment should lead.
When students understand models, data, prompts, retrieval, and agents, they can build more thoughtful AI projects instead of only copying tool tutorials.
Each week pairs one core concept lesson with an interactive activity and a clear student takeaway.
What counts as AI; rules-based software vs. machine learning; narrow AI; where students encounter AI; history of AI.
AI Scavenger Hunt: students identify AI in apps they use, then classify each example as recommendation, prediction, generation, recognition, or automation.
AI is not just ChatGPT. It is a collection of technologies that recognize patterns, make predictions, and perform tasks.
Data, labels, features, patterns; training vs. inference; supervised learning; how dataset quality affects outputs.
Human Machine-Learning Game: teams receive example data and create rules for classifying new cases, such as spam vs. non-spam messages.
AI learns patterns from examples rather than being explicitly programmed for every situation.
Generative AI; large language models; tokens; next-token prediction; probability; context windows; why outputs vary.
Next-Word Predictor: students predict the next word in sentences, compare probabilities, then give the same prompt to an AI multiple times and compare answers.
Language models generate likely continuations; they do not retrieve a perfectly stored answer every time.
Context, role, task, constraints, examples, output format; iterative prompting; prompt limitations.
Prompt Makeover Challenge: students improve a vague prompt over several rounds and compete to produce the clearest study guide, story, or product idea.
Better instructions usually produce better results, but prompting cannot eliminate every AI limitation.
Hallucinations; factuality vs. plausibility; fake citations; source checking; evaluation criteria; basic AI evals.
AI Fact-Checker: teams ask AI difficult or obscure questions, verify the answers using reliable sources, and score the responses for accuracy and confidence.
AI can sound confident while being wrong, so important claims must be verified.
Bias in data; fairness; privacy; personal information; copyright; deepfakes; misinformation; responsible use in school.
AI Ethics Court: groups analyze cases involving facial recognition, AI hiring, deepfakes, or student monitoring and argue what should be allowed.
AI decisions affect real people, and technical performance is not the only measure of whether a system is good.
Models that understand or generate text, images, speech, music, and video; diffusion explained simply; synthetic media.
One Idea, Four Formats: students turn one concept into a written description, generated image, short audio script, and video storyboard.
Modern AI can work across multiple forms of information, not just text.
Model vs. product; APIs; system prompts; retrieval-augmented generation; embeddings explained conceptually; connecting AI to documents.
Build a Mini Knowledge Assistant: students organize a small set of class documents and design how an AI assistant should find and use the correct information. No coding required.
Useful AI products usually combine a model with data, instructions, interfaces, and other software.
Chatbots vs. agents; planning; tools; memory; multi-step tasks; human approval; MCP as a connection standard.
Human Agent Simulation: one student acts as the AI planner, while others act as tools such as search, calculator, calendar, and documents to complete a complex task.
Agents do more than respond: they plan, use tools, observe results, and continue toward a goal.
Different models and providers; speed, quality, cost, context, privacy, reasoning, and multimodal capabilities; open vs. closed models.
Model Olympics: students test the same tasks across two or three available models and create a scorecard for accuracy, creativity, speed, and usefulness.
There is no universally best AI model; the right choice depends on the task and constraints.
Students can join weekly office hours for help, review, and optional deeper exploration.
Students get weekly classes, office hours, and a practical foundation for building with AI responsibly.