AI & LLM Glossary
A searchable glossary of 50+ AI and LLM terms — tokens, embeddings, RAG, temperature, transformers — in plain English. Free, in your browser.
AI & LLM Glossary
Loading terms…
No terms match your search. Try a shorter or different keyword.
How to Use the AI & LLM Glossary
Browse or search
The glossary opens with every term listed alphabetically. Scroll to browse, or start typing in the search box to jump straight to what you need.
Search term or meaning
The filter matches both the term name and its definition. Type "search", for example, and you'll see semantic search, vector database and reranking, not just one entry.
Read the plain-English definition
Every entry is one or two clear sentences with no jargon-by-jargon explanations. The running count shows how many terms match your current search.
Use it as you read or build
Keep it open beside an AI article, a model's docs, or your own project. It runs entirely in your browser, so it's fast and private — nothing you type is sent anywhere.
A Shared Vocabulary Is the Fastest Way Into AI
Why the words matter as much as the technology
Artificial intelligence has moved from research labs into everyday tools faster than its vocabulary has spread. Open any product announcement, model card, or tutorial and you hit a wall of terms — tokens, embeddings, temperature, RAG, fine-tuning, context windows — each assumed to be already understood. The result is a strange gap: people use these systems daily but can't always say what is happening under the hood, which makes it hard to use them well, compare them fairly, or talk about them clearly. A glossary closes that gap. When everyone in a conversation means the same thing by "context window" or "hallucination", discussion stops being a guessing game and starts being useful.
The words also carry real consequences. If you don't know that a knowledge cutoff exists, you'll trust a model on yesterday's news and be misled. If you don't know what temperature does, you'll blame the model for being either too repetitive or too unpredictable when a single setting was the cause. If you've never met prompt injection, you might paste untrusted content into an agent and hand an attacker the keys. Understanding the vocabulary isn't academic — it is the difference between using AI confidently and being surprised by it. Each definition here is deliberately short and concrete, written so that someone meeting the term for the first time walks away able to use it in a sentence correctly.
"You can't reason about a system whose words you don't share. The vocabulary is the on-ramp — once the terms click, the technology stops feeling like magic and starts feeling like engineering."
How to get the most from this glossary
Treat the search box as the main entrance. The filter matches both the term and its definition, so a single keyword surfaces a whole cluster of related ideas: search for "vector" and you'll find embeddings, cosine similarity, semantic search and vector databases sitting together, which is exactly how they work in practice. That makes the glossary useful for more than a one-off lookup — it's a map of how the concepts connect. Reading three or four neighbouring entries in one sitting often teaches more than memorising a single definition, because most AI ideas only make sense in relation to the others around them.
The terms here span the whole stack, from the lowest-level mechanics to the highest-level patterns. There are the building blocks — tokens, parameters, weights, attention, transformers. There are the controls you actually touch — temperature, top-p, top-k, max tokens, stop sequences. There are the training ideas — pre-training, fine-tuning, LoRA, RLHF, distillation, quantization. And there are the application patterns that dominate today's products — RAG, agents, tool calling, MCP, structured output and grounding. Whether you're a curious reader, a writer covering AI, a student, or a developer shipping a feature, the goal is the same: give you the words so the ideas stop being intimidating. Bookmark it, keep it open while you work, and come back whenever a new term shows up in your reading.
10 Facts About AI Terminology
A token isn't a word — it's a word-piece, so "tokenization" itself can split into several tokens.
The word "transformer" comes from the 2017 paper "Attention Is All You Need", which underpins almost every modern LLM.
"Hallucination" is a metaphor — the model isn't confused, it's predicting plausible text without a notion of truth.
Temperature borrows its name from physics: higher "heat" means more random, varied output.
Embeddings turn meaning into geometry — similar ideas literally sit closer together in space.
RAG stands for Retrieval-Augmented Generation, a phrase coined in a 2020 research paper.
Top-p sampling is also called "nucleus sampling" — two names for the very same setting.
A model's parameter count (billions of weights) is only a rough proxy for how capable it is.
Few-shot, one-shot and zero-shot simply count how many examples you put in the prompt.
This glossary runs entirely in your browser — your searches are never uploaded.
Frequently Asked Questions
- It's a searchable reference of more than 50 of the most common artificial-intelligence and large-language-model terms, each explained in one or two plain-English sentences. It's designed for quick lookups while you read, study or build with AI.
- Type into the search box at the top. The filter matches both the term name and the text of the definition, so a single keyword like "vector" surfaces every related entry. A live count shows how many terms match.
- No. The entire glossary is bundled into the page and filtered with plain JavaScript in your browser. Nothing you type is sent to a server, a model, or any third party, and nothing is saved.
- No. The definitions are a fixed, hand-written list. There's no model running at all — the tool is fully deterministic and works offline once the page has loaded.
- Anyone meeting AI terms for the first time or wanting a quick refresher — curious readers, writers, students, product managers and developers alike. The definitions assume no prior background.
- A token is the unit a model actually processes, often a word-piece or a few characters rather than a whole word. Long or rare words split into several tokens, which is why models measure length and cost in tokens, not words.
- Temperature controls how much randomness a model uses when picking the next token. Low temperature gives safe, predictable, repetitive output; high temperature gives more varied and creative — but less reliable — output.
- Retrieval-Augmented Generation means fetching relevant documents and feeding them to the model so its answer is grounded in specific, up-to-date information instead of only what it memorised during training.
- No — the terms are vendor-neutral. Concepts like tokens, embeddings, context windows and temperature apply across ChatGPT, Claude, Gemini and open models alike, though exact behaviour can vary by provider.
- Completely free, with no account, sign-up or usage limit. It runs in your browser and collects no data.
Related News
You may be interested in these recent stories from our newsroom.
No related news yet for this tool. Our editorial team publishes new pieces every week.
Browse all news →75 more free tools
Calculators, converters, security tools — no signup.