מילון AI
כל מונח AI, מוסבר
44 מונחים חיוניים בעולם ה-AI — מ-LLM ועד RAG ו-MCP — בעברית פשוטה. שמרו את הדף; אנחנו מעדכנים את ההגדרות ככל שהתחום מתפתח.
AI Agent
An AI system that uses LLMs not just to chat, but to plan, take actions (call APIs, run code, browse the web), and pursue multi-step goals autonomously.
AI Alignment
The discipline of ensuring AI systems behave in ways that match human values and intentions — both in safety (don't cause harm) and in usefulness.
AI Art
Images, illustrations, animations, or designs generated primarily by AI tools — the output of Stable Diffusion, Midjourney, DALL-E, and similar generators.
Anthropic API
Programmatic access to Anthropic's Claude models with prompt caching, extended thinking, computer use, file analysis, and the Agent SDK.
Chain-of-Thought (CoT) Prompting
Asking an LLM to 'think step by step' before answering — significantly improves performance on reasoning tasks.
Chatbot
An AI-powered conversational interface — ChatGPT, Claude, Gemini, customer support bots, and assistants — built on LLMs in modern usage.
ChatGPT
OpenAI's flagship chatbot, launched November 2022. The product that brought conversational AI to the mainstream — now serving 200M+ weekly active users.
Claude
Anthropic's chatbot, often considered the strongest LLM for reasoning, code, and long-form writing in 2026. Built with a constitutional AI safety approach.
Context Window
The maximum number of tokens an LLM can process in one interaction — including your prompt, conversation history, and the model's response.
Cursor
AI-first code editor (a VS Code fork) with chat, autocomplete, and an agent mode that edits multiple files. The default editor for AI-assisted coding in 2026.
DALL-E
OpenAI's image generation model, integrated into ChatGPT. Known for the best prompt adherence in the category.
Diffusion Model
The AI architecture behind most image generators (Stable Diffusion, DALL-E, Midjourney) — generates images by progressively denoising random noise.
ElevenLabs
The leading AI voice synthesis platform — natural narration, voice cloning from 30 seconds of source audio, 29+ languages.
Embedding
A vector representation of text, image, or audio — a list of numbers that captures the semantic meaning, enabling 'find similar' searches.
Few-shot Prompting
Showing an LLM 1-5 examples of input-output pairs in the prompt, before asking it to handle a new input — dramatically improves accuracy.
Fine-tuning
Taking a pre-trained AI model and continuing to train it on your specific data so it specializes for your use case (medical, legal, customer support style, etc.).
Gemini
Google's flagship LLM family. Distinguishing feature: 2M-token context window — large enough to load entire codebases or books.
GitHub Copilot
Microsoft / GitHub's AI coding assistant — autocomplete in IDEs, chat in VS Code, agent mode via Copilot Workspace. The pioneer of AI coding.
Hallucination
When an LLM confidently produces information that's factually wrong — invented citations, fake quotes, plausible-sounding but incorrect claims.
Image Generation
AI that creates new images from text descriptions, reference images, or both — covering tools like Midjourney, DALL-E, Stable Diffusion, and Adobe Firefly.
Inference
The act of running a trained AI model to generate output — i.e., 'using' the model, as opposed to training it.
LLM (Large Language Model)
An AI system trained on massive text datasets to predict and generate human-like text — the technology behind ChatGPT, Claude, Gemini, and most modern AI chatbots.
LoRA (Low-Rank Adaptation)
An efficient fine-tuning technique that trains a small 'adapter' on top of a frozen base model — fast to train, tiny to store, and stackable.
MCP (Model Context Protocol)
An open standard launched by Anthropic in 2024 that defines how AI models communicate with external tools and data sources.
Midjourney
AI image generator known for the highest aesthetic quality in the category. Originally Discord-based, now also web-based.
Multimodal
An AI model that handles more than one type of input — text + images + audio + video — typically in the same prompt.
OpenAI API
Programmatic access to OpenAI's models (GPT-5, o-series, embeddings, DALL-E, Whisper, Sora). The most widely-used commercial LLM API.
Perplexity
Conversational AI search engine — every answer cites real-time web sources. The 'AI search' alternative to ChatGPT.
Prompt Engineering
The practice of crafting input text (prompts) to get the best output from an LLM. A vague prompt gives vague answers; a structured prompt gives precise ones.
RAG (Retrieval-Augmented Generation)
A technique that lets an LLM answer questions about information it wasn't trained on — by retrieving relevant documents at query time and stuffing them into the prompt.
RLHF (Reinforcement Learning from Human Feedback)
The technique that turns a base LLM into a useful assistant — by having humans rate model responses and using that feedback to fine-tune behavior.
Runway
AI-first video editing platform combining generative video with a full editing suite — masks, motion brush, timeline, camera controls.
Sora
OpenAI's text-to-video model, launched publicly in early 2024. Industry-leading raw video quality, 60-second clips on the Pro tier.
Stable Diffusion
The most influential open-source image generation model, released by Stability AI in 2022 — the foundation of much of the AI art ecosystem.
Suno
AI music generator producing full songs with vocals from text prompts. The leading consumer music AI in 2026.
System Prompt
A high-priority instruction given to an LLM at the start of a conversation — defines the assistant's role, constraints, and persona.
Temperature
A parameter that controls how random an LLM's output is — 0 = deterministic and safe, 1 = creative and varied, >1 = chaotic.
Token
The basic unit that LLMs read and produce. Roughly 0.75 words in English. APIs charge per token consumed and produced.
Transformer
The neural network architecture underlying modern LLMs and most image AI — introduced by Google in 2017 and quickly the dominant approach.
TTS (Text-to-Speech)
AI that converts written text into spoken audio — modern TTS (ElevenLabs, OpenAI Voice, Hume) is indistinguishable from human voice in most use cases.
Vector Database
A database optimized for storing and searching embeddings — finding the nearest vectors to a query vector in milliseconds, even across billions of records.
Video Generation
AI that creates video clips from text prompts, reference images, or existing video — Sora, Runway, Veo, Kling, and Pika lead the category.
Voice Cloning
AI that copies a specific person's voice from a short sample (30 seconds to a few minutes), then generates new speech in that voice.
Zero-shot
Asking an LLM to do a task with no examples in the prompt — relying purely on the model's pre-trained knowledge.