The Little AI Lexicon: 20 Terms Finally Clear

Prompt, token, hallucination, LLM, RAG, fine-tuning: conversations about AI sometimes resemble a dinner where everyone speaks a foreign language — and where you nod hoping no one asks a question. Good news: this intimidating vocabulary covers simple ideas, explained in a few sentences with the right images. Here’s the lexicon of the 20 terms that pop up everywhere — finally clear, to follow conversations, read articles, and understand what you’re already using.

The fundamentals: the machine and its learning

Artificial Intelligence (AI): an umbrella term — any program performing tasks that would require human intelligence (understanding, recognizing, deciding). Machine Learning: the go-to method — instead of programming rules, you show millions of examples and the machine extracts patterns; like learning to recognize a cat by seeing a thousand cats, not by reading its definition. Neural Network: the architecture loosely inspired by the brain — millions of tiny connected computing units that adjust during learning. Deep Learning: neural networks with many layers — the recipe behind the spectacular progress of the decade. Model: the result of learning — the « brain » trained for use; when someone says « a new model is out, » understand « a new version of the brain. »

Generative AI and its stars

Generative AI: the one that creates — texts, images, sounds, videos — instead of just classifying or recognizing. LLM (Large Language Model): the engine behind conversational assistants — a giant model trained on oceans of text, becoming a language expert. Prompt: your instruction — the request you formulate; the art of crafting it well is called prompt engineering. Token: the text’s unit of division (a short word, a piece of a long word) — AI limits and costs are measured in tokens; remember « piece of a word. » Hallucination: the famous flaw — the AI confidently states something false, as it optimizes plausibility, not truth; hence the rule to always verify the important. Multimodal: understanding and producing multiple « modes » — text, image, sound — in the same conversation.

The right reflex. To anchor these concepts, ask an AI itself: « explain [term] with a daily analogy, then a concrete example. » The AI vocabulary explained by AI, at your pace and with your images: it’s the perfect exercise to combine the useful with the metalinguistic — and check that you now know how to dialogue with it.

Under the hood: the technical terms that keep coming up

Training: the long and costly phase where the model learns from data. Inference: the usage phase — when the trained model responds to your requests; each answer is an inference. Fine-tuning: specializing a generalist model in a domain — like sending a good generalist to expert training. RAG: connecting AI to reference documents it consults before answering — the open-book exam, which reduces hallucinations and allows answering on your content. Parameters: the billions of internal settings in a model — the order of magnitude cited to give its « size. » Open source / open weights: models freely downloadable and usable on your own machines — as opposed to closed models accessible only via their publisher.

The ecosystem and the debates

Agent: an AI that no longer just responds but chains actions (search, use tools, execute) to accomplish a mission — the current big trend. Bias: the prejudices inherited from training data — the AI reflects what it was shown, flaws included. AGI (Artificial General Intelligence): the hypothetical AI matching humans in all domains — the subject of heated debates and no current reality; when you read AGI, understand « speculation. » Alignment: the field working to ensure AI does what we expect, safely and in line with our values — increasingly important as they gain autonomy.

Frequent questions

Do you need to know these terms to use AI?

No — you can use an assistant just fine without vocabulary. But knowing them helps understand the limits (hallucination, bias), read the news without being misled, and choose your tools knowingly.

What’s the difference between ChatGPT, an LLM, and AI?

Simple nesting: AI is the entire field; the LLM is a language-specialized model; known assistants are products built around an LLM. Like « transport, » « gas engine, » and « a car brand. »

Will the vocabulary still change?

Yes — each wave brings its words (the « agents » still confidential yesterday are everywhere). But the fundamentals of this page remain stable: they’ll serve as your durable foundation.

What to remember

AI jargon covers simple ideas: the machine learns by example (machine learning, neural networks), assistants rely on giant language models (LLMs) guided by instructions (prompts), their flaws have names (hallucinations, biases), and their evolutions too (agents, multimodal, RAG). With these twenty terms, you follow conversations, decode announcements, and spot the fog sellers. And remember the best AI vocabulary teacher: AI itself — an analogy and an example on demand, until it’s clear.