AI glossary

Key AI terms in plain language

12 terms

  • LLM

    Large language model — a network trained to predict the next text token (GPT, Claude, Gemini).

  • Токен

    A piece of a word (~0.75 English words, ~0.5 Ukrainian), the unit for API volume and cost.

  • Контекстне вікно

    How many tokens the model sees at once (8K–1M). Beyond the window, it forgets.

  • Промпт

    Instruction to the model: role + context + task + output format.

  • Галюцинація

    Confidently generated falsehood: invented facts, sources, links.

  • Файнтюнінг

    Further training on your data for a stable style/format.

  • RAG

    Retrieval-augmented generation: the model searches your documents and answers with citations.

  • Ембединг

    Vector representation of text for semantic search (distance = similarity).

  • Агент

    Model + tools + loop: plans and executes steps itself (function calling).

  • Temperature

    Generation randomness: 0 deterministic, 1+ creative.

  • System prompt

    Hidden developer instruction shaping assistant behavior.

  • Hallucination guard

    Anti-fabrication techniques: source citations, 'answer only from data', verification.