Video summary

🤖Qué es un Prompt en Inteligencia Artificial y Para qué sirve: EJEMPLOS Reales

Main summary

Key takeaways

Educational

Main ideas / lessons

  • What a prompt is (in AI): A prompt is an input—text, image, or even sound—that you provide to an AI model (e.g., language models) so it can understand your intention and generate the output you want.

  • Prompt outputs depend on the tool/model: Depending on the AI system, prompts can produce text, images/graphics, sound, or other formats.

  • Prompts work better with context + clear instructions: The more you specify the role, audience, format, constraints, and topic, the better the results.

  • Avoid ambiguous requests: If your prompt is unclear, the AI may make incorrect assumptions (e.g., interpreting “teach me to paint” as painting a picture rather than painting home walls).

  • Long instructions are harder to process at once: Many tools struggle with overly long or information-dense prompts in a single interaction. It’s recommended to:

    • break tasks into successive prompts
    • ask the AI to confirm understanding before proceeding

  • You often need to refine the first answer: The initial response is rarely perfect; request edits, corrections, or additional details.

Methodology / process: 4 phases of how a prompt works

The video describes a 4-step workflow for prompting:

  1. Interpret the instruction

    • The AI begins by understanding what you asked.
    • Interpretation happens because the model has limits and can’t absorb everything perfectly at once.

  2. Generate a response

    • The AI produces an output based on its interpretation.

  3. Adjust / refine the response

    • You may need to ask for modifications or add more constraints.
    • Example refinement: rewrite headlines with specific brand constraints.

  4. (Most neglected) where mistakes often happen

    • The final phase is frequently ignored and tied to the corrections/refinement loop—verifying the output truly matches your intent and requirements.

Examples shown (and what they demonstrate)

1) Text prompt example (ChatGPT): roles + brand voice + audience

  • Task: Write a winter beard-care article for the hipster sector.
  • Prompt strategy demonstrated:
    • Assign a role: “content manager for a cosmetics store specializing in beard care products”
    • Assign a personality/brand voice: Harley Davidson
    • Specify topic: “how to care for your beard in winter”
    • Ask for confirmation: “Answer yes or no” (confirm understanding)
  • What it shows: Rich context + explicit instruction yields on-target results quickly.

Then the demo asks for output ideas:

  • Request five alternative headlines
  • The AI returns multiple title options within milliseconds.

2) Image prompt examples (Midjourney)

  • Purpose: Show that prompts also work for image generation.

  • Examples demonstrated:

    • A short instruction-like prompt that yields four images (Midjourney commonly produces 4 variations)
    • A more structured prompt referencing style/subject and movie inspirations (e.g., futuristic android robot, transparent elements, face handling)
    • A longer prompt written in natural language with a scenario/shot idea (e.g., “I imagine Adam walking…”), producing impressionistic-style images

  • What it shows: Image tools can respond to both:

    • compact “keyword” style prompts
    • longer natural-language descriptions

3) Warning example — ambiguity leads to wrong interpretation

  • Prompt: “Teach me to paint.”
  • Problem shown: The AI assumes “paint a picture” (recreational painting) instead of painting the walls of the home.
  • Lesson: Be explicit about what painting means (walls vs artwork).

4) Warning example — missing product details harms an ad

  • Prompt: “I want to sell shower screens, please write an ad for Instagram.”
  • Problem shown: The AI doesn’t get key details (screens are for shower/bath, not generic “in general” items).
  • Lesson: Include essential attributes (context, use-case, target customer/product specifics).

5) Refinement example — maintain style without mentioning a brand

  • After receiving headlines in a Harley-like voice, the next prompt refines constraints:
    • “Rewrite the headlines without mentioning the Harley Davidson brand, but maintaining its voice and personality.”
  • What it shows: You can preserve tone while changing constraints.

Key recommendations

  • Use rich context (role, audience, tone/voice, topic, required format).
  • Confirm understanding early (e.g., ask yes/no).
  • Prefer multiple shorter prompts over one huge prompt.
  • Make instructions unambiguous (avoid leaving critical details implied).
  • Iterate: revise outputs until they match your intent and constraints.
  • For GPT-like tools: prompts can be questions, phrases, sentences, or paragraphs.
  • For tools with built-in GPT versions (mentioned):
    • there are default specialized versions (e.g., browser, games/video games tutor, negotiator)
    • you can use the “classic” version with your own instructions and/or create tailored GPTs
  • Marketing note in the video: The creator encourages downloading free resources from a “school” and subscribing/liking/sharing—this is promotional rather than instructional.

Speakers / sources featured

  • Speaker: Unspecified video creator/instructor (first-person narration; no name provided).
  • Tools/brands referenced as sources in examples:
    • ChatGPT / OpenAI GPT
    • Midjourney
    • DALL·E (mentioned as integrated with GPT for image creation)
    • Harley Davidson (example brand voice in prompts)
    • Guggenheim Museum (image-generation target)
    • Bilbao (location used in the image-generation example)
    • Adam (character used in a natural-language image prompt example)

Original video