The 212AY Library
AI
Glossary.
200+ artificial intelligence terms explained in plain language — from machine learning and LLMs to AI agents, RAG and prompt engineering. Every definition comes with a concrete example, in English, French and Arabic.
200 terms
Intent Recognition
Intent recognition is the NLP task of identifying what a user wants to accomplish based on their text or speech input. When a customer types «I want to change my flight», the system classifies the intent as a flight modification request and may extract entities like date and flight number. It is the backbone of conversational AI: virtual assistants, customer service bots, and voice-controlled systems all rely on accurate intent recognition to route requests to the right action.
Interpretability
Interpretability is the degree to which humans can understand how an AI model reaches its decisions, by examining its internal mechanisms rather than just its outputs. It matters most in high-stakes settings: if a bank's model refuses a small business loan in Rabat, regulators and customers deserve to know which factors weighed in. Deep neural networks are often «black boxes», so researchers develop techniques to inspect what happens inside them. Interpretability builds trust, reveals hidden biases, and enables meaningful audits.
Jailbreaking
Jailbreaking is the practice of crafting prompts that trick an AI model into bypassing its safety rules and producing content it was designed to refuse, such as harmful instructions or confidential data. Attackers use role-play scenarios, encoded messages, or elaborate hypotheticals to fool the model. For a business deploying a customer-facing chatbot, jailbreaking is a real risk: a manipulated bot could insult clients, reveal internal policies, or promise illegal discounts. Regular red-teaming and layered guardrails are the standard defenses.
JSON Mode
JSON mode is a setting offered by many LLM APIs that forces the model to reply in valid JSON, the structured format software systems exchange. Instead of a free-form paragraph, the model returns organized fields that code can reliably parse. For example, an e-commerce platform in Casablanca can send customer emails to a model in JSON mode and receive back fields like intent, product, and urgency, ready to route into its CRM automatically.
Jupyter Notebook
Jupyter Notebook is an open-source tool that lets you mix executable code, results, charts, and explanatory text in a single interactive document. It is the standard workbench for data scientists: you run code cell by cell, see outputs immediately, and document your reasoning alongside. A data analyst at a Moroccan telecom operator might use a notebook to explore churn data, visualize trends, and share the annotated analysis with managers. Google Colab offers the same experience free in the browser.
K-Nearest Neighbors (KNN)
K-nearest neighbors is a simple machine learning algorithm that classifies a new data point by looking at the k most similar examples already seen and adopting their majority label. No real training happens: the algorithm just memorizes the data and compares distances. Think of pricing a riad in Marrakech by checking the five most similar riads recently sold nearby and averaging their prices. KNN is intuitive and effective on small datasets, but slows down considerably as data volume grows.
Kaggle
Kaggle is a Google-owned online platform where data scientists find datasets, share notebooks, take free courses, and compete in machine learning challenges with real prizes. It is one of the best places to learn AI by doing: a student in Fès can practice on real-world data, study winning solutions from top practitioners, and build a public portfolio that employers recognize worldwide.
Knowledge Distillation
Knowledge distillation is a technique where a large, powerful «teacher» model trains a smaller «student» model to imitate its behavior, transferring most of the capability at a fraction of the size and cost. It is like a master artisan in Fès training an apprentice: the apprentice never sees every piece the master ever made, but learns the craft by observing the master's work. Distilled models run cheaper and faster, making AI viable on modest servers or mobile phones.
Knowledge Graph
A knowledge graph is a structured representation of information as a network of entities and the relationships between them: people, products, places, and companies connected by labeled links. Where a database stores rows, a knowledge graph stores facts like «Company X supplies Product Y to Client Z». A Moroccan bank can use one to map connections between accounts and companies to spot fraud rings, and combining knowledge graphs with LLMs helps ground answers in verified facts rather than guesses.