AI Bookmarks Dashboard
← Back to themes
Prompting

Prompting

5 videos

Prompt engineering techniques, prompt libraries, wording and structure tips.

  1. Loop Engineering vs. Prompt Engineering — The Doctor's Second Opinion Analogy

    Conceptual only — no code shown. The implementation detail (how the second opinion is structured in a loop) is covered in follow-up videos by this creator.

    After generating your main output, add: 'Now act as a reviewer. Read the output above and check whether it fully and correctly addresses the original request. If anything is missing, incorrect, or could be improved, say so explicitly. Otherwise confirm it's complete.'

    View full video →
  2. Building a Loop in Claude.ai: Two Scheduled Tasks That Take Turns (Sourcer + Checker)
    1. In Claude.ai, go to Scheduled Tasks and create a new task manually.
    2. Set Task 1 ('Sourcer'): runs hourly, prompt instructs agent to generate output and write to a shared store (e.g. Notion database).
    3. Create Task 2 ('Checker'): runs hourly, prompt instructs agent to read the shared store and validate/critique each item.
    4. Optional: add Task 3 to validate the checker's output.
    5. Review the decision log in Notion to see the loop's output accumulate over time.

    Create two Claude.ai scheduled tasks that alternate hourly. Sourcer prompt: 'Search for [topic] and write each result with key details to [shared doc/Notion database].' Checker prompt: 'Read each entry in [shared doc] and for each one, assess whether it meets [criteria]. Mark it as Valid/Invalid and explain why.'

    View full video →
  3. Use Claude to Convert Text Explanations Into Readable HTML Pages
    1. Ask Claude to explain a tool or concept you want to learn.
    2. When Claude returns a blob of text, follow up: "Convert that explanation to HTML."
    3. Open the HTML in a browser — you get a structured page instead of a wall of text.
    View full video →
  4. Stop Describing Vibes to AI — Use Design Style Names Instead
    1. When prompting AI for design work:
    2. Replace vague descriptors with named design styles: instead of 'clean' say 'Japandi'; instead of 'functional and no nonsense' say 'Utilitarian'; instead of 'soft and bookish' say 'Light Academia'.
    3. Find an article with 50+ design style names — each style describes the look, mood, and appropriate use case.
    4. Mix styles for unique results: e.g. 'Light Academia meets Scrapbook', 'Neo-brutalism meets Minimalist'.
    5. Apply this to any AI design task: UI, landing pages, graphics, presentations.

    Design [component] in a [style name] aesthetic (e.g. Utilitarian, Japandi, Neo-brutalism, Light Academia). Apply the full look and mood of that style — not a generic interpretation.

    View full video →
  5. How Robinhood Serves Live Stock Prices to Millions (Pub-Sub & Conflation)

    None found — conceptual system design explainer, no step-by-step implementation instructions.

    View full video →