An AI Glossary: What All Those Terms You Hear Every Day Actually Mean

When the talk turns to artificial intelligence, strange words inevitably appear: hallucinations, fine-tuning, embeddings, transformer. People nod along, but few understand what is meant. It is time to fix that.

Why we need a glossary now

Over the past two years, AI has moved out of niche technical circles into the mainstream. Everyone talks about ChatGPT, Claude and Gemini, but the underlying terminology has stayed the same: technical, tangled and full of anglicisms. People hear the words in podcasts, see them in articles, hear them at conferences and at work, yet don't really grasp their meaning.

The problem is not that people are dim. It is that the AI community is used to speaking its own language, built on decades of research. It is time to translate these ideas into plain Russian: not marketing gloss, but real, working explanations that will keep you from looking lost in a conversation.

The main categories of AI terms

AI terms fall loosely into several groups, and each group has its own logic. If you understand the logic of one group, the other terms will fall into place naturally.

Models and architectures are the foundational concepts of how a neural network itself is built: transformer, diffusion model, large language model (LLM), convolutional neural network. They are like architectural styles for a building: each determines how everything else will be arranged.

Training and adaptation covers the processes a model goes through in order to learn: fine-tuning, prompt engineering, reinforcement learning, backpropagation. Think of it as an athlete's training regimen: it determines which skills the model will acquire.

Problems and limitations are what developers and users actually run into: hallucination, bias, alignment, overfitting, mode collapse. They are like illnesses a system can have, and they need to be taken into account.

From hallucinations to prompt engineering

Let's take the five most important and most frequently used terms:

• Hallucinations: when an AI generates information that is not in its training data. It invents links, quotes and figures that never existed. It sounds like mystical fantasy, but it is actually an error in text prediction.
• Fine-tuning: additional training of an already finished model on specialized data. It is as if you took a ready-made textbook and adapted it for a specific school.
• Embeddings: converting words or text into mathematical vectors that the model can understand. It is a translation of human language into the language of numbers.
• Transformer: the network architecture underlying all modern LLMs (GPT, Claude, Llama). It was a breakthrough in 2017 that changed the entire AI world.
• Prompt engineering: the art of phrasing a task for an AI correctly so as to get the result you need. Not just a command, but a command that works.

Why it matters

A glossary is not just a reference book. It is a pass into serious conversations about AI. Once you understand that a hallucination is a specific, explainable error in a model's behavior, the conversations change entirely. You no longer nod blindly at conferences; you ask the right questions. At work, you don't just listen to colleagues, you take part in the discussion.

Understanding the terminology is the first step toward understanding what AI can and cannot do, why it works, and why it sometimes breaks.