Generative AI Explorer
You have spent eight lessons building a deep understanding of how generative AI works — how text is generated token by token, how images emerge from noise, how voices are synthesized, how multimodal systems integrate different formats, what these systems do well, and how to use them responsibly. Now it is time to apply that knowledge hands-on. This lesson is structured as a guided exploration: you will interact with generative AI tools (or, if tools are not available, reason through the activity using what you know), then analyze and reflect on what you observe.
Before You Begin: Your Evaluation Framework
Good exploration requires a framework — a consistent set of questions you bring to every test so your observations are comparable and analytical rather than random. As you work through each activity below, keep these four questions in mind for every output you examine: Accuracy: Is the content factually correct, or can you spot errors, outdated information, or hallucinations? Coherence: Does the output make sense as a whole? Does it stay on topic, maintain consistent logic, and flow naturally? Prompt fidelity: Did the AI actually do what you asked? Did it respect your constraints — length, tone, format, subject? Unexpected failures: Did the AI produce anything strange, wrong, or concerning that you did not expect?
When you interact with an AI tool, you are not just a user — you are a researcher. Record what you asked, what you got, and how it compares to your predictions. The gap between prediction and result is where the most interesting learning happens.
Generative AI Exploration Lab
- PART 1 — TEXT GENERATION EXPERIMENT
- Choose a text-based generative AI tool you have access to (for example, Claude, ChatGPT, Gemini, or Copilot). If none is available, proceed to the analysis questions using what you know from the module.
- Round A — Factual Test: Ask the AI a factual question about a historical event you know well. Record the exact prompt and the full response. Evaluate: Is every claim accurate? Are there any subtle errors? Now ask the same question with the instruction 'include three specific citations.' Examine the citations carefully — do they appear to be real? How would you verify them?
- Round B — Creative Test: Prompt the AI to write a three-paragraph short story set on Mars, told from the perspective of a robot geologist who has just discovered evidence of ancient water. Evaluate the output using all four framework questions. Then change one element of the prompt and generate again. Note what changed in the output.
- Round C — Limit Test: Ask the AI a question about an event from the past week. Record the response. Does the model acknowledge its knowledge cutoff? Does it guess? Does it hallucinate? What does this reveal about the model's self-awareness of its own limits?
- PART 2 — IMAGE GENERATION EXPERIMENT (if a tool is available)
- Use a free image generation tool (Adobe Firefly, Microsoft Designer, Canva AI, or similar).
- Round A — Specificity Test: Generate an image from a vague prompt ('a city'). Then generate from a highly specific version of the same prompt ('a cyberpunk city at dusk, neon reflections on rain-soaked cobblestones, no people, wide-angle shot, digital painting style'). Compare the two images. Write a paragraph on how specificity affected the result.
- Round B — Limit Test: Ask the generator to include readable text in the image — for example, a shop sign saying 'OPEN' in large letters. Examine the result carefully. Is the text legible? What does this reveal about how the model handles symbolic information versus visual patterns?
- PART 3 — COMPARISON AND REFLECTION
- After completing Parts 1 and 2, answer these questions in writing:
- 1. Which modality — text or image generation — surprised you most? Why?
- 2. Describe one specific example of a strength you observed and one specific example of a limit or failure you observed. Connect each to the concepts from earlier lessons.
- 3. If you were explaining to a younger sibling what generative AI is good for and what it is bad at, based only on what you observed today, what would you say?
- 4. Write one question about generative AI that these experiments raised that you do not yet know the answer to.
Reflecting on the Experience
Hands-on exploration of AI tools builds a kind of intuition that reading about them cannot. You begin to develop a feel for where the models shine, where they struggle, and how your choice of prompt changes the outcome. That intuition is valuable — but it needs to be disciplined by the analytical framework you brought to the exploration. The students who use generative AI most effectively are not the ones who are most impressed by it or most suspicious of it. They are the ones who know what questions to ask of every output, who verify what matters, and who treat AI as a capable but fallible partner rather than an oracle.
Hands-on experience builds intuition for where AI succeeds and fails. The analytical framework — accuracy, coherence, prompt fidelity, unexpected failures — turns that intuition into reliable judgment. Both are needed.
Flashcards — click each card to reveal the answer
When testing a language model with a factual question, a student notices the answer contains one detail that is subtly wrong. What should they conclude?
Why does asking an AI image generator to put readable text in an image usually fail?