INDEX

1.- INTRODUCTION

2.- Learning Objectives (Copilot)

3.- What you will learn (Google Gemini)

4.- How the AI models are trained?

5.- Generative AI vs Traditional AI

6.- How to ask for a drawing to Gemini                                                                       

7.- Elements of AI

7.1.- How should we define AI?

INTRODUCTION.-

Artificial intelligence is neither science fiction nor a distant concept; it is an invisible tool that is already a core part of our daily decisions. From selecting the fastest route on a map to discovering new content on social media, AI acts as a silent assistant, simplifying our routines almost without us noticing. 

The true revolution lies in how machines learn. Unlike traditional programming, where a human dictated every instruction step-by-step, modern AI functions through observation. By processing massive volumes of data (such as examples of music or images), the system can independently identify the characteristics that define each element, 'learning' to distinguish complex concepts without the need for rigid mathematical definitions.

AI functions through patterns, writing its own rules based on what it observes

The primary purpose of this technology is to act as an enhancer of our own capabilities. By delegating mechanical tasks or dense data analysis to AI, we gain time and clarity. This not only makes us more productive in the workplace but improves our quality of life by allowing us to focus our energy on higher-value tasks.

As a leading example of this technology, Google introduces Gemini, an assistant designed to fuse technical power with ease of use. Its tools allow you to go far beyond simple text: it can conduct Deep Research, generate advanced visual content (Nano Banana), or digitise personal files (such as family recipes) and analyse videos. It is, in essence, an engine for creativity and organisation.

Gemini no busca en Internet en tiempo real, sino que sus respuestas están basadas en la gigantesca biblioteca que ya tiene memorizada.

---

The true revolution lies in how machines learn. Unlike traditional programming, where a human dictated every instruction step-by-step, modern AI functions through observation. By processing massive volumes of data (such as examples of music or images), the system can independently identify the characteristics that define each element, "learning" to distinguish complex concepts without the need for rigid mathematical definitions.

AI functions through patterns, writing its own rules based on what it observes.

Back to the index

---

Learning Objectives (Copilot).-

• Examine strategies for writing articles, developing creative ideas, and understanding how Copilot can facilitate the generation of automated responses to improve customer interaction.
• Explore how to useMicrosoftCopilot to interact and generate visual content, applying the tool in creating images for social media, presentations, and more.
• Discover the ethical use of Microsoft Copilot, emphasizing the importance of data privacy and security, and how to safely manage access and passwords.

What is an artificial intelligence assistant?

Think of an AI assistant as a sophisticated digital collaborator that uses specialized software to simulate human conversation and perform tasks. Unlike older programs that follow a rigid "if this, then that" script, an AI assistant uses Machine Learning to understand context, learn from interactions, and handle complex requests.

Core Components.-

To function effectively, these assistants rely on several key technologies:

• Natural Language Processing (NLP):- This allows the AI to parse human speech or text, understanding not just the words, but the intent and sentiment behind them.
• Large Language Models (LLMs):- These are the "brains" (like the model I am based on) trained on vast amounts of data to predict and generate human-like responses.
• Automation Hooks:- The ability to "talk" to other apps, allowing the assistant to set timers, send emails, or control smart home devices.

Key Capabilities.-

Modern AI assistants generally fall into three categories based on what they do best:

Category	Primary Use Case	Examples
Generative	Creating content, coding, and complex reasoning.	Gemini, ChatGPT
Utility/Voice	Hands-free tasks, reminders, and smart home control.	Alexa, Siri
Enterprise	Specific business workflows, data analysis, and customer support.	Salesforce Einstein, GitHub Copilot

What Makes Them Unique?

Unlike a standard search engine that gives you a list of links, an AI assistant synthesizes information.

1. Context Awareness: They remember what you said earlier in the conversation to provide relevant follow-up answers.
2. Multimodality: Many modern assistants can 'see' images, 'hear' audio, and 'speak' back to you in real-time.
3. Creative Problem Solving: They can brainstorm ideas, write poetry, or debug a broken line of code rather than just retrieving a pre-written answer.

A Quick Reality Check.-

It is important to remember that while I can sound very human, I don't have feelings, personal beliefs, or a physical existence. I process patterns in data to be as helpful as possible.

Back to the index

---

How the AI models are trained?

Training an AI model like me is a massive, multi-stage process that transforms raw data into an assistant capable of reasoning. It is often compared to a student going through school: first learning to read, then following instructions, and finally learning how to behave politely in society.

Here is the breakdown of the three main stages of modern AI training:

1. Pre-training: 'The Knowledge Phase'
   This is the most resource-intensive stage. The model is fed trillions of words from the internet, books, and code.
   • The Goal: To learn the 'statistical structure' of language.
   • How it works: The model plays a high-stakes game of "fill in the blank." It hides a word in a sentence and tries to guess it. If it’s wrong, it adjusts its internal parameters (billions of them) to be more accurate next time.
   • Result: The model becomes a walking encyclopedia, but it doesn't know how to "talk" yet—if you ask it a question, it might just give you more questions because that’s what it saw on the web.
2. Supervised Fine-Tuning (SFT): 'The Instruction Phase'.-
   Once the model understands language, humans step in to teach it how to be an assistant.
   • * **The Goal:** To teach the model how to follow commands (e.g., 'Summarize this' or 'Write a Python script').
   • How it works:** Trainers provide thousands of 'Prompt-Response' pairs. For example:
     • Prompt: 'What is the capital of France?'
     • Ideal Response: 'The capital of France is Paris.'
   • Result:** The model learns the format of a conversation and starts acting like a helpful assistant.
3. Reinforcement Learning from Human Feedback (RLHF): 'The Alignment Phase'.
   This is the 'polish' stage that makes an AI safe and helpful.
   • The Goal:** To align the AI’s behavior with human values (being honest, harmless, and helpful).
   • How it works:** 1. The model generates several different answers to the same prompt.
     2. A human 'ranker' looks at them and picks the best one.
     3. A separate Reward Model is trained to predict what humans like.
     4. The AI is then 'rewarded' for generating answers that the Reward Model likes.
   • Result:** The AI learns to avoid toxic content, admit when it doesn't know an answer, and match the user's tone.

Training Summary Table.-

Stage	Data Source	Analogy	Key Outcome
Pre-training	Massive Web/Book Scrapes	Reading every book in a library.	World knowledge & Grammar.
SFT	Expert-written Q&A	Taking a "How to be an Assistant" course.	Ability to follow instructions.
RLHF	Human rankings/preferences	Learning etiquette and safety from a mentor.	Safety, nuance, and personality.

The "Reasoning" Trend (2025–2026).-

Recently, a new stage called Reasoning Fine-Tuning has emerged (used in models like DeepSeek or OpenAI's 'o' series). Instead of just predicting the next word, the model is trained to generate a 'Chain of Thought'—showing its work step-by-step before giving the final answer. This is done using reinforcement learning where the model is rewarded specifically for getting the correct answer in math or logic.

Back to the index

---

What you will learn (Google Gemini).-

• Mastering prompting¹Prompting is essentially the art of guiding the model to give us the best possible response. And the key for this to work well is to provide good context.: Apply the four pillars (specificity, context, format, and examples) to achieve brilliant results.
• Research & learning: Utilise tools such as Deep Research and NotebookLM to analyse documents and videos, and generate detailed reports in minutes.
• Boosting creativity & productivity: Integrate Gemini into the Google ecosystem to manage calendars and summarise documents.
  Create stunning visual content with tools like Nano Banana and Veo, and use Canvas to build interactive presentations and code.

USE DELIMITATORS IN YOUR PROMPTS: Just as people understand text better with line breaks, separators, quotation marks ..., AI also benefits from having a clear structure

Back to the index

---

Generative AI vs Traditional AI.-

The main difference between Generative AI and Traditional AI comes down to their core mission: one is a creator, and the other is a judge.

While traditional AI is built to analyze and categorize existing data, generative AI is designed to use its training to produce brand-new content.

Note: Traditional AI excels in accuracy and reliability, while generative AI excels in creativity and adaptability. However, generative AI can experience 'hallucinations' (inventing data), something that rarely occurs with traditional AI.

1. The Core Purpose.-

• Traditional AI (Discriminative/Predictive): Focuses on pattern recognition to make a decision or a prediction. It looks at data and asks: "What is this?" or "What will happen next based on these rules?"

• Generative AI: Focuses on creation. It learns the underlying structure of data so it can ask: "Can I make something new that looks like the data I've seen?"

2. Quick Comparison Table.-

Feature	Traditional AI	Generative AI
Output	A label, a number, or a "Yes/No".	Text, images, audio, video, or code.
Logic	Evaluates and discriminates.	Generates and synthesizes.
Example Task	Identifying a face in a photo.	Creating a new face that doesn't exist.
Goal	Accuracy and classification.	Innovation and simulation.
Well-known Apps	Google Maps, Netflix Recs, Siri.	ChatGPT, Midjourney, Gemini.

3. A Simple Analogy: The Art Gallery.-

Imagine both types of AI are standing in an art gallery:

• Traditional AI acts like a Curator. You show it a painting, and it can tell you instantly: "This is a 19th-century Impressionist oil painting of a lily." It is excellent at sorting and finding things.

• Generative AI acts like the Artist. You tell it: "Paint me a 19th-century Impressionist lily," and it picks up a brush to create a version that has never been seen before.

---

4. How They Work Together.-

In the real world, these two often work as a team. For example, in a modern self-driving car:

1. Traditional AI identifies the stop signs, pedestrians, and lane lines (Classification)

2. Generative AI can be used in the lab to simulate millions of "fake" weather scenarios or traffic accidents to help the car learn how to react without needing a real-world crash (Simulation).

One thing to keep in mind...

Traditional AI is generally more deterministic (it gives the same answer to the same data). Generative AI is probabilistic, meaning it might give you a slightly different (and sometimes "hallucinated" or incorrect) answer every time you ask.

Back to the index

---

La Inteligencia Artificial Generativa (IA generativa) es una rama de la inteligencia artificial que se centra en crear contenido nuevo en lugar de simplemente analizar o clasificar datos existentes.

A diferencia de la IA tradicional (que podría decirte si una foto contiene un perro), la IA generativa puede dibujar el perro, escribir un cuento sobre él o incluso componerle una canción desde cero.

¿Cómo funciona?

Se basa en modelos de aprendizaje profundo (Deep Learning) que son entrenados con conjuntos masivos de datos (textos, imágenes, código). El sistema aprende los patrones, estructuras y estilos de esos datos para generar resultados que son originales pero coherentes.

Capacidades principales.-

• Generación de texto: Creación de artículos, correos, poemas o código de programación (ej. ChatGPT, Gemini).
• Creación de imágenes: Transformación de descripciones textuales (prompts) en arte visual o fotografías (ej. Midjourney, DALL-E).
• Audio y Vídeo: Composición de música, clonación de voz o creación de escenas de vídeo realistas.
• Resolución de problemas: Ayuda en el diseño de nuevas moléculas para medicinas o materiales industriales.

Diferencia clave.-

IA Tradicional (Predictiva/Analítica)	IA Generativa
Analiza datos para encontrar patrones	Utiliza patrones para crear algo nuevo
Elige una opción entre varias (Clasifica)	Inventa una respuesta (Crea)
Ejemplo: Filtro de spam en el correo	Ejemplo: Escribir el correo por ti

Es una herramienta poderosa que está transformando industrias creativas y técnicas, aunque también plantea retos importantes sobre la ética y la veracidad de lo que produce.

Back to the index

---

General Prompts for Copilot.-

These instructions cover a variety of possible uses for artificial intelligence, ranging from designing detailed plans to interpreting text and behaviour, adopting specific personalities, translating languages and justifying strategic decisions. Each type of instruction exploits different capabilities of AI, allowing users to apply it in educational, professional, entertainment or analytical contexts.

1.- Explain.-

This type of instruction asks the agent to detail and clarify a concept, process or idea in an understandable way. It is useful for learning about complex topics or understanding something better. Example: "Explain how an internal combustion engine works" or “Explain how artificial intelligence can change the approach to education in the coming years”.

2.- Describe.-

To detail the visual, sensory, or descriptive characteristics of an object, place, person, or situation. Also the characteristics of an idea. This instruction is used to get a clear picture of something without seeing it or to understand a concept in detail. Example: “Describe the appearance of a tropical forest” or “Describe the changes that AI is introducing in medicine”. 

3.- Summarise.-

This prompt requires the virtual agent to condense a larger amount of information into a shorter version, keeping the key points. It is very useful for getting the gist²the gist of: La esencia de of long texts or speeches. Example: “Summarise the main arguments of the book ‘1984’ by George Orwell” or “Summarise the most important events that led to the fall of the Berlin Wall”.

4.- Elaborate.-

With this instruction, we ask Microsoft Copilot to generate specific content, whether it be a blog article on a specific topic, a social media post to promote an event, or a summary of a technical report.

5.- Compare.-

Comparing and contrasting is an instruction that allows two or more elements to be requested and evaluated, highlighting their similarities and differences. It helps to better understand the relationships and distinctive features between concepts. Example: “Compare and contrast market economies and planned economies” or “Compare and contrast the ideas of Socrates and Plato”.

6.- Generate an Idea.-

Back to the index

---

How to ask for a drawing to Gemini

Para obtener ese aspecto específico que ves en tus imágenes, te recomiendo pedir un estilo de **ilustración editorial 2.5D** o **ilustración digital con sombreado suave (soft shading)**.

Tus dibujos no son planos (2D puro como *Los Simpson*), pero tampoco son modelos tridimensionales complejos (3D como *Toy Story*). Se encuentran en ese punto intermedio donde el dibujo es lineal pero tiene **volumen y profundidad**.

Aquí tienes los términos clave que puedes usar en tu próximo "prompt" para mantener la consistencia:

### 1. Estilo de Sombreado y Volumen
* **Sombreado suave (Soft Shading):** Esto evita los cortes bruscos de sombra y da esa apariencia redondeada a las caras y la ropa.
* **Iluminación volumétrica:** Para que los personajes parezcan tener cuerpo y peso dentro del escenario.
* **Gradientes sutiles:** Ayudan a que las superficies no parezcan planas.

### 2. Definición del Trazo
* **Línea de contorno definida (Clean Line Art):** Los personajes están claramente delimitados por una línea fina y oscura, lo cual es típico del dibujo animado clásico.
* **Grosor de línea variable:** Ayuda a acentuar la profundidad.

### 3. El Género Visual
* **Estilo "Corporate Memphis" evolucionado:** Es un estilo moderno de ilustración digital muy usado en apps y artículos, pero con más detalle y realismo en las proporciones.
* **Iluminación ambiental:** Fíjate cómo en la imagen del restaurante hay focos de luz cálida; pedir "iluminación ambiental cálida" ayuda a recrear esa atmósfera.

---

Ejemplo de cómo podrías pedírmelo:

Dibuja una escena en estilo de ilustración digital 2.5D, con personajes de trazo limpio y sombreado suave para darles volumen. La estética debe ser similar a una caricatura editorial moderna, con colores vibrantes y una perspectiva que dé profundidad al escenario.

Ejemplo.-

Se le pide a Gemini que haga el siguiente dibujo, tanto en 2,5 D como en 3 D.:

Haz un dibujo en estilo de ilustración digital 2.5D, con personajes de trazo limpio y sombreado suave para darles volumen. La estética debe ser similar a una caricatura editorial moderna, con colores vibrantes y una perspectiva que dé profundidad al escenario/ /El dibujo será de una playa solitaria y, entre dos palmeras cuelga una hamaca. En esta hamaca está tumbado un hombre/ /El fichero adjunto es un modelo para la cara del hombre/ /El formato del dibujo debe ser apaisado, esto es, más ancho que alto. Ancho=1013 px. Alto=530 px/

Los resultados son:

Back to the index

---

Elements of AI.-

How should we define AI?

Implications of AI's applications.-

Implications for self-driving cars: road safety should eventually improve as the reliability of the systems surpasses human level. The efficiency of logistics chains when moving goods should improve. Humans move into a supervisory role, keeping an eye on what’s going on while machines take care of the driving. Since transportation is such a crucial element in our daily life, it is likely that there are also some implications that we haven’t even thought about yet.

Implications for content recommendation: while many companies don’t want to reveal the details of their algorithms, being aware of the basic principles helps you understand the potential implications: these involve so called filter bubbles, echo-chambers, troll factories, fake news, and new forms of propaganda.

Implications for image and video processing: when such techniques advanced and became more widely available, it got much easier to create natural looking fake videos of events that are impossible to distinguish from real footage. This challenges the notion that “seeing is believing”.

What is, and what isn’t AI? Not an easy question!

The popularity of AI in the media is in part due to the fact that people have started using the term when they refer to things that used to be called by other names.

That is because of:

1. No officially agreed definition.- Even AI researchers have no exact definition of AI. The field is rather being constantly redefined when some topics are classified as non-AI, and new topics emerge.
2. The legacy of science fiction.- The confusion about the meaning of AI is made worse by the visions of AI present in various literary and cinematic works of science fiction.
3. What seems easy is actually hard...- Another source of difficulty in understanding AI is that it is hard to know which tasks are easy and which ones are hard.
   For instance grasping objects: While easy for you, by a robot is extremely hard, and it is an area of active study.
4. ...and what seems hard is actually easy.- By contrast, the tasks of playing chess and solving mathematical exercises can seem to be very difficult.
   Both tasks follow a set of rules and are very easy to computers, able to compute billions of computations a second.

So what would be a more useful definition?

An attempt at a definition more useful than the “what computers can’t do yet” joke would be to list properties that are characteristic to AI, in this case autonomy⁴The ability to perform tasks in complex environments without constant guidance by a user. and adaptivity⁵The ability to improve performance by learning from experience..

Back to the index