Decode LLM, RAG, RLHF and other AI abbreviations in minutes. This compact glossary provides practical definitions for developers, investors, and product teams.
Artificial intelligence is rapidly changing the world, and at the same time creating a new language to describe how it works. Abbreviations such as LLM, RAG, and RLHF pop up frequently at modern product conferences, pitches, and panels, which can confuse even the most experienced professionals. This Ukrainian AI glossary provides a compact and practical reference to the terms you are most likely to come across when developing, investing, or simply reading about artificial intelligence. This field is rapidly evolving, so we will update it regularly.
Key concepts in the AI glossary
A.G.I. – General artificial intelligence. This concept typically refers to systems that can perform a wide range of tasks at or above the human level. Although different formulations are provided by different sources, the general consensus is that AGI includes the ability to handle most cognitive tasks.
AI agent – Autonomous systems that use AI to perform a series of tasks from the user. Such agents can act on several AI systems, such as expense reporting, reservations, and code creation, although the exact meaning of the term may vary depending on context and available infrastructure.
API endpoint – API endpoints that serve as interfaces for program-to-program integration. This allows one program to fetch data from another program or manage third-party services without manual intervention. As AI agents develop, these endpoints will become more widely used.
chain of thoughts – Chain of Reasoning: Break down a task into intermediate steps step by step. For large language models, this kind of thinking improves the accuracy of the answer, but requires more time to justify each step. In model training, this is used to improve the quality of the conclusions through special methods.
computing – Compute power required to train and deploy models. This is mainly hardware. Graphics processing units (GPUs), central processing units (CPUs), tensor processing units (TPUs), and other infrastructure that form the backbone of the modern AI industry.
deep learning – A subfield of machine learning that uses multilayer artificial neural networks. Such architectures can discover complex relationships in the data, but require large amounts of data and long training times compared to simpler methods.
Large-scale language model (LLM) – Large language models, neural networks with billions of parameters. Process natural language and generate text responses. Examples include ChatGPT, Claude, Google Gemini, Meta Llama, Microsoft Copilot, and Mistral Le Chat.
neural network – A multi-layered architecture that powers neural networks, deep learning and modern generative technologies. Use interactions between layers to transform input data into useful signals.
open source – Open source software. Such an approach allows researchers and companies to collaborate to develop and analyze solutions, providing transparency and auditability. Closed solutions have limited access to internal logic.
reinforcement learning – Reinforcement learning: The agent learns by interacting with the environment, receives rewards for correct actions, and gradually shapes its behavior. In modern models, this is often combined with human feedback (RLHF) to make responses more useful and safe.
token – The basic unit of data in human-model interaction. Tokens break text into smaller fragments for processing by the model. The cost of using LLM is often calculated by the number of tokens.
training – The process of training a model on large amounts of data to detect patterns and form useful responses. Training can be resource-intensive and may be supplemented by techniques such as fine-tuning to adapt to specific tasks.
The AI glossary is updated regularly as the field is rapidly evolving. It will be a useful reference for developers, investors, and anyone following trends in artificial intelligence.
