The AI Dictionary I Wish I Had When I Started Learning Artificial Intelligence

The Problem No One Talks About in AI Courses

You open an exciting Artificial Intelligence course, full of expectations. Within the first ten minutes, the instructor has already thrown out terms like tokens, embeddings, fine-tuning, neural networks, RAG, and temperature. Everyone in the room seems to understand. You, don't.

Relax. The problem is not you. The problem is that no one teaches the language of AI before teaching AI itself.

It’s like trying to learn medicine without knowing what "diagnosis" or "anatomy" mean. The feeling is exactly that: an invisible wall made of technical jargon that keeps away those who need to learn the most.

It was with this realization that student and developer Meghana Reddy created her AI Dictionary — a guide published on Medium that went viral among technology students worldwide. The proposal is simple and powerful: to explain fundamental AI concepts in accessible language, with everyday examples and easy-to-remember analogies.

In this article, we adapt and expand that material for the Brazilian context, bringing the essential vocabulary that every beginner in AI needs to master.

---

1. Artificial Intelligence (AI)

Simple definition: It’s the ability of machines to perform tasks that normally require human intelligence.

Technical definition: A branch of computer science focused on building systems capable of reasoning, learning, solving problems, understanding language, and making decisions.

Everyday example: Google Maps calculating the fastest route in real time.

AI example: ChatGPT answering questions in natural language.

Common uses:

• Virtual assistants (Alexa, Siri, Google Assistant)
• Recommendation systems (Netflix, Spotify)
• Fraud detection in banking
• Autonomous cars
• Medical imaging diagnosis

Common mistake: Confusing AI with Machine Learning. AI is the broad field; Machine Learning is one of its ways to achieve it.

---

2. Machine Learning

Simple definition: It’s the process of teaching computers to learn from data, without needing to manually program each rule.

Analogy: Imagine teaching a child to recognize cats. You don’t list all the technical characteristics of a cat — you show hundreds of pictures until they learn on their own. Machine Learning works this way.

Practical example: A spam filter that learns which emails are unwanted based on your behavior over time.

Main types:

• Supervised Learning: The model learns from labeled data (e.g., cat photo = "cat")
• Unsupervised Learning: The model finds patterns on its own, without labels
• Reinforcement Learning: The model learns through trial and error, receiving rewards for correct actions

---

3. Deep Learning

Simple definition: It’s an advanced type of Machine Learning that uses neural networks with many layers to process complex data.

Why "deep"? Because the data passes through various layers of processing — each layer learns something more sophisticated than the previous one.

Where it's used:

• Facial recognition
• Machine translation
• Image generation by AI
• Medical imaging diagnosis

Analogy: Think of the human brain. When you see a face, neurons in different layers process colors, shapes, textures, and ultimately, the identity of the person. Deep Learning imitates this process.

---

4. Neural Network

Simple definition: It’s a computational system inspired by the functioning of the human brain, composed of interconnected nodes (artificial neurons) organized in layers.

Basic structure:

• Input Layer: Receives the raw data
• Hidden Layers: Process and transform the data
• Output Layer: Produces the final result

Example: A neural network trained to identify tumors in X-rays analyzes pixels, then textures, then shapes, and finally classifies the image as "tumor" or "healthy".

---

5. Token

Simple definition: It’s the smallest unit of text that an AI model processes. It can be a word, part of a word, or even a character.

Practical example: The phrase "Artificial Intelligence" can be divided into tokens like: ["Artificial", " Intellig", "ence"] — depending on the model.

Why it matters: Language models like GPT-4 have a limit of tokens they can process at once (called the context window). This affects how much text you can send in one conversation.

---

6. Embedding

Simple definition: It’s a way of representing words, phrases, or images as numbers (vectors) such that similar concepts are mathematically close together.

Analogy: Imagine a map where nearby cities have similar coordinates. Embeddings do the same with meanings: "king" and "queen" are close in vector space, just like "dog" and "hound".

Why it’s powerful: It allows the AI to understand that "automobile" and "car" mean the same thing, even though they are different words.

---

7. Prompt

Simple definition: It’s the instruction or question you give to an AI model to obtain a response.

Bad example: "Write something about marketing."

Good example: "Write a persuasive sales email, in an informal tone, for a young audience aged 18 to 25, introducing a personal finance app."

Prompt Engineering is the art of creating effective prompts — a skill increasingly valued in the job market.

---

8. Fine-Tuning

Simple definition: It’s the process of taking an already trained AI model and retraining it with specific data for a particular task or domain.

Analogy: Imagine a general practitioner who specializes in cardiology. He already knows medicine; fine-tuning is this specialization.

When to use: When you want the AI to respond with the specific language, tone, and knowledge of your business or field.

---

9. RAG (Retrieval-Augmented Generation)

Simple definition: It’s a technique that combines a language model with an external database. Instead of relying solely on what it learned during training, the model fetches updated information before responding.

Analogy: It’s like the difference between a student who only uses what they've memorized (pure model) and one who can consult books during an exam (RAG).

Why it’s important: It solves the issue of models that become "outdated" or "make up" answers (a phenomenon called hallucination).

---

10. Temperature

Simple definition: It's a parameter that controls how much "creativity" or randomness an AI puts into its responses.

| Temperature | Behavior | |---|---| | Low (0.1 – 0.3) | More precise, predictable, and conservative responses | | Medium (0.5 – 0.7) | Balance between precision and creativity | | High (0.9 – 1.0) | More creative, varied, and sometimes unpredictable responses |

When to use low temperature: Data analysis, coding, technical responses.

When to use high temperature: Creative writing, brainstorming, idea generation.

---

11. LLM (Large Language Model)

Simple definition: These are AI models trained on huge amounts of text to understand and generate human language.

Notable examples:

• GPT-4 (OpenAI)
• Gemini (Google)
• Claude (Anthropic)
• LLaMA (Meta)

What they do: Answer questions, write texts, translate languages, generate code, summarize documents, and much more.

---

12. Hallucination

Simple definition: When an AI model generates false or invented information with complete confidence as if it were true.

Real example: Asking ChatGPT for bibliographic references and it invents scientific papers with plausible titles, authors, and dates — but that never existed.

How to avoid: Always verify critical information from external sources. Use techniques like RAG for models in production.

---

Why Learning This Vocabulary Changes Everything

Knowing these terms is not just an academic issue. It’s what separates those who consume AI from those who use it strategically.

When you understand what temperature is, you know how to adjust your prompts. When you understand embeddings, you can build semantic search systems. When you understand RAG, you know how to create reliable corporate chatbots.

The AI vocabulary is the key to transforming confusion into clarity — and clarity into professional opportunity.

---

Conclusion: Start with the Language

As Meghana Reddy rightly observed in her original article, the biggest obstacle in learning AI is not the mathematical complexity — it’s the barrier of jargon. When you decode the language, the path becomes infinitely more accessible.

Keep this article as a reference. Return whenever you encounter an unknown term. And remember: every expert was once a lost beginner amid technical terms.

The difference is that now you have the dictionary they didn’t have.