AI glossary
AI terms in three languages, in plain words. Every term has its own link you can share.
57 terms
Agent skill
Agent koʻnikmasi (skill) / Навык агента (skill)
A folder of instructions (a SKILL.md file) that teaches an AI agent to do a particular task the same way every time.
AI agent
AI agent / ИИ-агент
A model that completes a task in several steps on its own: it plans, uses tools and checks the result.
Alignment
Moslashtirish (alignment) / Согласование (alignment)
Work to make a model act helpfully, safely and in line with what people intend.
API
API / API
The interface programs use to talk to each other. Through an AI API you plug a model into your own app.
Artificial intelligence (AI)
Sunʼiy intellekt (AI) / Искусственный интеллект (ИИ)
Computer systems that do tasks that usually need human intelligence: understanding text, recognising images, making decisions.
Attention
Diqqat mexanizmi (attention) / Механизм внимания
The mechanism inside a transformer that decides which other words in the text matter for each word.
Automation
Avtomatlashtirish / Автоматизация
Handing repetitive work to software or AI: sorting emails, preparing reports.
Benchmark
Benchmark / Бенчмарк
A standard test for comparing models on the same tasks.
Bias
Ogʻish (bias) / Предвзятость
Unfair answers towards certain groups caused by imbalances in the training data.
Chatbot
Chatbot / Чат-бот
A program that answers questions in a conversation. Modern chatbots run on LLMs.
Classification
Tasniflash / Классификация
Sorting data into set groups: is an email spam or not, is a review positive or negative.
Computer vision
Kompyuter koʻrishi / Компьютерное зрение
AI that understands images and video: recognising faces, products, reading documents.
Context window
Kontekst oynasi / Контекстное окно
How much text a model can 'see' at once: your question, the chat history and any uploaded documents together.
Copilot
Yordamchi (copilot) / Копайлот (ассистент)
An AI assistant built into your work: it suggests as you write code, documents or email.
Data labeling
Belgilash (labeling) / Разметка данных
Adding the correct answer to examples: tagging a photo 'cat', marking an email 'spam'.
Dataset
Dataset (maʼlumotlar toʻplami) / Датасет
A collection of examples gathered to train or test a model.
Deep learning
Chuqur oʻqitish / Глубокое обучение
Machine learning with many-layered neural networks. Today's chat models and image generators are built on it.
Deepfake
Dipfeyk / Дипфейк
A fake video, photo or voice made with AI. Scammers use them, so be careful.
Diffusion model
Diffuziya modeli / Диффузионная модель
A model that creates an image by removing noise step by step. Stable Diffusion and Midjourney work this way.
Embedding
Embedding (vektor koʻrinish) / Эмбеддинг
Turning the meaning of text into a list of numbers. Texts with similar meaning get similar numbers, which powers search.
Fine-tuning
Qayta oʻqitish (fine-tuning) / Дообучение (fine-tuning)
Training an existing model further on your own data to adapt it to a specific task.
Generative AI
Generativ AI / Генеративный ИИ
AI that creates new content: text, images, audio, video or code.
GPT
GPT / GPT
Generative Pre-trained Transformer. The name of OpenAI's models and a general type of model.
GPU
GPU (videokarta) / GPU (видеокарта)
A processor that runs many calculations in parallel. The main hardware for training and running AI models.
Hallucination
Gallyutsinatsiya / Галлюцинация
When a model confidently states something wrong or made up. Always check important facts.
Inference
Inferens (xulosa chiqarish) / Инференс
Using an already-trained model: you send an input and get an answer.
Large language model (LLM)
Katta til modeli (LLM) / Большая языковая модель (LLM)
A model trained on huge amounts of text that writes by predicting the next word. ChatGPT, Claude and Gemini are LLMs.
Local model
Lokal model / Локальная модель
A model that runs on your own computer or server without the internet. Your data stays with you.
Machine learning
Mashinali oʻqitish / Машинное обучение
The branch of AI where a program learns from examples instead of rules written by a person.
MCP (Model Context Protocol)
MCP (Model Context Protocol) / MCP (Model Context Protocol)
An open standard for connecting AI assistants to files, apps and services.
MCP server
MCP server / MCP-сервер
A small add-on that gives an AI assistant new actions, such as reading Notion pages or opening a website in a browser.
Model
Model / Модель
A program that has learned from data and responds to new inputs, such as a language model answering a question with text.
Multimodal
Multimodal / Мультимодальный
A model that works with several kinds of data: text, images, audio, video.
Natural language processing (NLP)
Tabiiy tilni qayta ishlash (NLP) / Обработка естественного языка (NLP)
The field of AI that understands and produces human language: translation, summaries, text classification.
Neural network
Neyron tarmoq / Нейронная сеть
A model made of many connected simple computing units (neurons). The strength of the connections is tuned during training.
OCR
OCR (matnni tanish) / OCR (распознавание текста)
Turning text in a photo or scan into editable text.
Open source
Ochiq kod / Открытый код
Software whose code is public, so anyone can read, change and share it.
Overfitting
Haddan tashqari moslashish / Переобучение
When a model memorises its training examples and does badly on new data.
Parameters
Parametrlar / Параметры
The numbers inside a model that are tuned during training. A '7B model' has 7 billion parameters.
Prompt
Prompt / Промпт
The instruction or question you give a model. A precise prompt gets a precise answer.
Prompt engineering
Prompt muhandisligi / Промпт-инжиниринг
The skill of writing and testing prompts so a model gives good results reliably.
Quantization
Kvantlash / Квантование
Storing a model's numbers in a shorter format so it is smaller and faster. Needed to run it on an ordinary computer.
RAG (retrieval-augmented generation)
RAG (qidiruv bilan generatsiya) / RAG (генерация с поиском)
Before answering, the model looks up relevant passages in your documents and bases its answer on them.
Reasoning model
Fikrlovchi model / Рассуждающая модель
A model that 'thinks' through a problem step by step before answering. Stronger at maths and complex tasks.
Regression
Regressiya / Регрессия
Predicting a number: a flat's price, tomorrow's sales.
Reinforcement learning
Mustahkamlovchi oʻqitish / Обучение с подкреплением
A model learns by acting and getting rewards or penalties. Also used to tune chat models with human feedback.
Speech recognition
Nutqni tanish / Распознавание речи
Turning spoken words into text. Whisper, for example, also understands Uzbek.
Supervised learning
Nazoratli oʻqitish / Обучение с учителем
Training on examples where the right answers are known.
System prompt
Tizim prompti / Системный промпт
A standing instruction at the start of a conversation: the model's role, tone and rules.
Temperature
Temperatura / Температура
How random the answer is. Low gives precise, repeatable output; high gives more creative, varied output.
Text-to-speech (TTS)
Matndan nutq (TTS) / Синтез речи (TTS)
Turning written text into a natural-sounding voice.
Token
Token / Токен
A small chunk of text a model reads and writes: a word, part of a word or a symbol. Prices and limits are usually counted in tokens.
Tool use
Vositalardan foydalanish / Использование инструментов
A model's ability to call outside functions: search, a calculator, a database, an API.
Training
Oʻqitish / Обучение
Showing a model many examples and adjusting its internal parameters so it answers correctly.
Transformer
Transformer / Трансформер
A neural network architecture introduced in 2017. Almost every modern LLM is built on it.
Vector database
Vektor maʼlumotlar bazasi / Векторная база данных
A database that stores embeddings and quickly finds the closest in meaning. A core part of RAG systems.
Weights
Ogʻirliklar (weights) / Веса
The values of a trained model's parameters. 'Open weights' means you can download them.