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
Semantic Search
Semantic search retrieves information based on meaning rather than exact keyword matching. Instead of matching the exact phrase, it understands that «affordable hotel near the medina» and «budget riad in the old city» are asking for the same thing. It uses embeddings to convert queries and documents into vectors, then finds the nearest vectors. Semantic search is the foundation of modern RAG systems and product search engines.
Sentiment Analysis
Sentiment analysis automatically determines the emotional tone of text — positive, negative, or neutral — at scale. Businesses use it to monitor brand reputation, analyze customer reviews, track social media reactions, and prioritize support tickets. A Moroccan airline could run sentiment analysis on thousands of Twitter mentions to instantly spot a surge in complaints after a delayed flight, triggering an automated alert to the PR team.
Softmax
Softmax is a mathematical function that converts a vector of raw scores (logits) into a probability distribution, where each value is between 0 and 1 and all values sum to 1. It is the standard final layer in classification neural networks, turning the model's raw outputs into interpretable class probabilities. For example, a model predicting whether an email is spam might output logits [2.1, 0.3, -1.5], which softmax converts to probabilities [0.88, 0.11, 0.01].
Speech Synthesis
Speech synthesis (text-to-speech or TTS) is the AI task of generating natural-sounding spoken audio from text input. Modern TTS systems use deep learning models — often based on transformers or diffusion models — to produce speech that is nearly indistinguishable from human voice. Applications include virtual assistants, audiobook narration, accessibility tools for visually impaired users, customer service phone systems, and content localization. ElevenLabs, OpenAI's TTS, and Bark are prominent examples.
Supervised Learning
Supervised learning is the branch of machine learning where a model learns from labeled examples — input-output pairs where the correct answer is provided — to make predictions on new, unseen data. The two main tasks are classification (predicting categories, like spam vs. not spam) and regression (predicting continuous values, like house prices). It is the most widely used ML paradigm: most business applications — credit scoring, demand forecasting, medical diagnosis — rely on supervised learning.
Synthetic Data
Synthetic data is artificially generated data that mimics the statistical properties of real data without containing actual personal or sensitive information. It is used to augment small datasets, protect privacy, test systems, and train models when real data is scarce or regulated. A Moroccan hospital with few pediatric X-rays could generate synthetic X-ray images to train a diagnostic model without risking patient privacy.
t-SNE
t-SNE (t-Distributed Stochastic Neighbor Embedding) is a dimensionality reduction technique specifically designed for visualizing high-dimensional data in 2D or 3D. It maps similar data points close together and dissimilar ones far apart, revealing cluster structure that would be invisible in raw high-dimensional space. t-SNE is widely used to visualize embeddings, explore dataset structure, and present model behavior. However, it is slow on large datasets and primarily for visualization — not for preprocessing before model training.
Tabular Data
Tabular data is data organized in rows and columns, like a spreadsheet or database table, where each row represents an observation and each column a feature. It is the most common data format in business: sales records, customer databases, financial transactions, and HR systems all produce tabular data. Traditional ML algorithms (random forests, XGBoost, gradient boosting) often outperform deep learning on tabular data, making feature engineering and model selection especially important.
Temperature
Temperature is a parameter that controls the randomness of a language model's output. At temperature 0, the model always picks the most likely next token (deterministic, repetitive). As temperature increases, the model assigns more probability to less likely tokens, producing more diverse, creative, but also more unpredictable output. A temperature of 0.7 is common for conversational AI; 0.2 for factual question answering; 1.0+ for creative writing. It is one of the most important settings in prompt engineering.