You Should Be Teaching AI

AI education is positioned as the fastest path for early adopters to turn hands-on prompting experience into real-world impact: people who have already spent hundreds of hours with LLMs can package that hard-won intuition into clear lessons that help the rest of the world become productive sooner. The core message is blunt—AI’s power is arriving faster than most people’s understanding, and that gap can be closed by teaching fundamentals, not just sharing prompts.

A three-lesson course structure is used as the proof of concept, built around the idea that tiny wording changes can produce wildly different outputs because prompts manipulate specific “primitives” inside language models. The curriculum starts with core vocabulary and first principles: rather than treating prompting as a bag of tricks, it links prompting to how LLMs work under the hood so learners can reason about why an answer appears in a given scenario. Each lesson follows a repeatable pattern—an explanatory paragraph, a vocabulary section, a one-page chart for quick reference, then hands-on worked examples, short tasks, quizzes to force application, and a wrap-up on common pitfalls.

The first lesson demonstrates the mechanics through a simple example (gravity) and then shows how adding constraints narrows the output distribution. It also emphasizes experimental discipline: only change one variable at a time so learners can identify what caused a shift in results. The lesson ends by warning against vague prompts and by prescribing specificity in format—down to the difference between a “brief story” and a “brief bedtime story.” As the course progresses, the same structure expands into more advanced sampling concepts such as temperature and top P sampling.

Building the course is described as a workflow that blends course-platform tools with multi-step AI prompting. Teachable is used to assemble lesson blocks (text and images, quizzes, open-ended questions) and to generate automatic lesson summaries. For drafting, a multi-role prompt is used with GPT5 thinking to produce rigorous, first-principles explanations plus definitions, connections, and worked examples—while also requesting a reasoning trace so the author can verify the logic. A second prompt then converts the gathered material into an industry-ready course outline, with the outline serving as a “spine” that can be fleshed out into modules.

The author then iterates: copy the AI output into the course, tweak for clarity and textbook-like structure, and add visual proof by screenshotting real LLM runs. When rewriting lesson two into the preferred format, ChatGPT is used to reduce manual work, though some parts (like JSON-heavy lab templates) are toned down to avoid confusing learners. Even course visuals are generated with Gemini 2.5 Flash Image Gen, then edited to remove branding.

Overall, the project frames AI course creation as achievable and scalable: three lessons reportedly take about 6–8 hours with AI assistance, and the resulting materials can be made free to learners. The broader takeaway is that teaching specific AI workflows and prompting techniques can become a stable, even profitable, way to contribute—turning personal expertise into accessible education for others who are only now catching up.