The AI World from Above — A Complete Map

Have you ever tried to learn something new, only to be bombarded with jargon everywhere you look? AI, Machine Learning, Deep Learning, Neural Network, LLM… each one seems like a separate universe. But it is not.

In this episode, we are going to look at the entire world of artificial intelligence from above. Like drawing a general map before starting a big project. The goal is that after reading this episode, whenever you hear an AI term, you will know where it fits on the map.

AI, ML, DL, LLM — How Are They Related?

Let me start with an analogy. Imagine you have a set of boxes, each fitting inside the other — like Russian Matryoshka dolls.

The largest box is Artificial Intelligence (AI). This is a very broad concept. Any system that exhibits “intelligent” behavior falls under the AI umbrella. Even a simple if/else statement controlling an enemy in a video game is technically a form of AI.

Inside this large box, there is a smaller box called Machine Learning (ML). The difference is that instead of you writing the rules, you give the system data and it discovers the rules itself. For example, instead of saying “if the email contains the word lottery, it is spam,” you show it thousands of spam and normal emails and it learns to recognize the pattern.

Inside the ML box, there is another box: Deep Learning (DL). This method uses artificial neural networks. The word “deep” refers to the number of layers in these networks. The more layers, the “deeper” the network and the more complex patterns it can learn.

And finally, inside Deep Learning, we reach Large Language Models (LLMs). These are extraordinarily large neural networks trained on massive volumes of text that can understand and generate human language.

So the relationship is:

AI (Artificial Intelligence)
  └── Machine Learning
        └── Deep Learning
              └── LLM (Large Language Model)

Three Methods of Machine Learning — Simple and Painless

Now that we understand what Machine Learning is, let us see how machines actually “learn.” There are three main methods.

1. Supervised Learning

Imagine teaching a child to recognize animals. You show pictures and say “this is a cat,” “this is a dog,” “this is a bird.” After many examples, the child can recognize new animals on their own.

Supervised Learning works exactly the same way. You have data with “labels.” The model learns from this labeled data and can then predict labels for new data.

Real-world example: Email spam filters. Thousands of emails labeled “spam” or “normal” are fed to the model, and it learns to classify new emails.

2. Unsupervised Learning

Now imagine you have a basket full of fruit and tell a child to “sort them.” You do not tell them the fruit names. The child groups them by color, size, and shape on their own.

In Unsupervised Learning, data has no labels. The model must find patterns and groupings on its own.

Real-world example: Customer segmentation. An online store feeds its purchase data to a model, and it automatically groups customers into categories like “sale hunters,” “loyal customers,” and “one-time buyers.”

3. Reinforcement Learning

This one is interesting. Imagine training a dog. When it does the right thing, you give it a treat. When it makes a mistake, no treat. The dog learns through trial and error which actions lead to rewards.

Reinforcement Learning works the same way. An agent is placed in an environment, makes decisions, and receives “rewards” or “penalties” based on outcomes. Through repetition, it learns to make the best decisions.

Real-world example: Chess. Google’s AlphaZero reached superhuman level by playing millions of chess games against itself — without ever reading a chess book.

Why Is Everyone Talking About LLMs Now?

A good question: Machine Learning has existed for years. Deep Learning made big strides after 2012. So why is everyone suddenly talking about AI?

The simple answer: Because LLMs speak human language.

Before LLMs, using AI required being a programmer or data scientist. You had to call specific APIs, convert data into special formats, and interpret outputs. But LLMs give you an interface that every human is familiar with: natural language. You talk to them in English or any language and get answers. For the first time in history, AI is usable by everyone — not just specialists.

Three key factors drove the LLM revolution:

1. Transformer Architecture — In 2017, Google published a paper titled “Attention Is All You Need” introducing the Transformer architecture, showing that understanding language does not require word-by-word processing — all words can be analyzed simultaneously.
2. Massive Data — The internet has produced trillions of words of text. Books, articles, forums, Wikipedia, code, conversations… all became training food for LLMs.
3. Computing Power — Modern GPUs enabled training models with billions of parameters. Without this hardware, building LLMs would be impossible.

A Hidden Advantage You Already Have

You might think “I am a web developer who has worked with PHP or JavaScript for years — what does AI have to do with me?” But the reality is you have a significant advantage over someone starting from scratch.

It is called Systems Thinking.

As an experienced developer, you are accustomed to thinking systematically. You understand how input gets processed and output is produced. You know what debugging means. You know how to break a big problem into smaller pieces.

These are exactly the skills needed to work with AI:

• Understanding Input/Output — You are used to working with APIs. An LLM is ultimately an API: text goes in, text comes out.
• Understanding Pipelines — You know how data flows from one stage to the next. AI data pipelines work the same way.
• Debugging — When LLM output is poor, you need to figure out where the problem lies. Bad prompt? Input data issue? Wrong model? This is the same debugging mindset you have used for years.
• Software Architecture — You know how to design large systems. Building AI-powered applications requires exactly this skill.
• Understanding Trade-offs — In web development, you learned to balance speed, cost, and quality. In AI, it is exactly the same — just different variables.

Glossary — Terms You Should Know

Before we move on to the next episodes, let us review some key terms. You do not need to memorize them — just read through so they are familiar when you encounter them later.

• Model — The result of training an ML algorithm on data. Like the brain of an AI system where learned information is stored.
• Parameter — Numbers the model learns during training. More parameters generally means a larger and more powerful model. GPT-4 has approximately 1.8 trillion parameters.
• Training — The process of teaching a model. You show it data and its parameters get adjusted.
• Inference — Using a trained model. When you ask ChatGPT a question, you are performing inference.
• Token — The unit of text processing in LLMs. Roughly every 4 English characters count as one token. We will explain this in detail in Episode 3.
• Prompt — The text you give to an LLM for it to generate a response.
• Fine-tuning — Retraining a pre-trained model on your specific data to make it better for your particular task.
• Embedding — Converting text (or any data) into numbers so computers can work with them. We will cover this in detail in Episode 3.
• Hallucination — When an LLM says something untrue but with high confidence. Like a friend who confidently shares wrong information.
• Context Window — The maximum amount of text an LLM can consider in one conversation. Like short-term memory.

Our Roadmap

In this series, we will progress step by step:

1. Episode 1 (this episode): A general map of the AI world
2. Episode 2: AI Hardware — Why GPU matters and what options you have
3. Episode 3: Inside an LLM — Token, Embedding, and Transformer
4. Following episodes: Prompt Engineering, RAG, Fine-tuning, building Agents, and hands-on projects

Each episode builds on the previous one. Every concept used has been explained beforehand.

Summary

In this episode, we learned that:

• AI is a broad concept with ML, DL, and LLM as nested subsets
• There are three main methods of machine learning: Supervised, Unsupervised, and Reinforcement Learning
• LLMs created a revolution thanks to three factors: Transformer architecture, massive data, and computing power
• As an experienced developer, your Systems Thinking skills give you a significant advantage

In the next episode, we will explore hardware. We want to understand why GPU is the hero of the AI world, why VRAM matters so much, and how you can work with AI without buying expensive hardware. Ready?