30 秒でわかる
Andrew Ng demonstrates how beginners can describe a birthday-card web app, run generated HTML, add features and troubleshoot. He then applies the same iterative workflow to a table-tennis game.
- 1Describe the appDescribe inputs, outputs and desired behavior.
- 2Be specificSpecific prompts make results more predictable, not identical.
- 3Run the pageSave the HTML and open it in a browser.
- 4Iterate on featuresAdd features in small steps and test each change.
- 5Report bugsDescribe observed failures and verify suggested fixes.
- 6Try a gameApply the process to a table-tennis game.
- 7Keep learningShare your work and keep learning.
ポイント
- Describe inputs, outputs and desired behavior.
- Specific prompts make results more predictable, not identical.
- Save the HTML and open it in a browser.
- Add features in small steps and test each change.
- Describe observed failures and verify suggested fixes.
- Apply the process to a table-tennis game.
- Share your work and keep learning.
チャプターごとに整えた全文を読む
A first app
Describe and generate a birthday-card page.
Andrew Ng introduces a beginner’s exercise in building a web app with AI, using a birthday-card generator. He demonstrates a workflow rather than promising that every generated application will work without testing.
The starting point is a prompt describing the desired page, its inputs and the output. Ng shows how to add requirements step by step or combine a known specification into a longer initial request.
Specific instructions tend to produce more predictable results than a vague request, although multiple runs can still differ. The goal is to communicate the important behavior and appearance, then inspect the result.
Run and improve
Open the file, add features and fix issues.
He demonstrates saving generated HTML and opening it in a browser. The HTML file is an example of a self-contained application; sharing or deploying other apps may involve additional steps.
After the basic app works, learners can add fields, change the design or request a button to copy a message. Small iterations make it easier to see which change produced a problem.
When a button fails, Ng describes the observed behavior to the AI and asks for a fix. He encourages trying the revised app again; generated fixes can also be wrong and need verification.
Build something else
Apply the workflow to a game.
He applies the same process to a table-tennis game: ask for an initial version, then add difficulty levels, scoring and visual improvements. The exercise illustrates a repeatable approach rather than a fixed implementation.
Ng suggests trying other playful applications and showing results to friends for feedback. He closes by encouraging learners to combine hands-on building with further study.
同じチャンネルの記事
1:01:33EN
30:23EN