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
Residual Network (ResNet)
A residual network (ResNet) is a deep neural network architecture that uses skip connections — adding the input of a layer directly to its output — to enable training of very deep networks (hundreds or thousands of layers). Without skip connections, deep networks suffer from vanishing gradients where signals weaken through many layers. ResNets solved this problem and enabled the deep learning revolution in computer vision. They remain foundational for image classification, object detection, and feature extraction backbones.
Responsible AI
Responsible AI is the discipline of designing, deploying, and governing AI systems so they are fair, transparent, safe, privacy-preserving, and accountable. It translates ethical principles into concrete practices: bias testing, documentation, human oversight, impact assessments, and clear recourse when the system errs.
Retrieval-Augmented Generation (RAG)
RAG connects a language model to your own documents: when a question arrives, the system first retrieves the most relevant passages from a knowledge base, then passes them to the model so it answers using real, up-to-date company information instead of only its training data. This sharply reduces hallucinations and keeps answers current without retraining.
Reinforcement Learning from Human Feedback (RLHF)
RLHF is the training technique that turned raw language models into helpful assistants like ChatGPT and Claude. Humans compare pairs of model answers and pick the better one; a reward model learns these preferences; then reinforcement learning optimizes the language model to produce responses humans prefer, making it more helpful, honest, and safe.
Robotic Process Automation (RPA)
Robotic Process Automation (RPA) uses software robots to mimic human actions on computer systems — clicking buttons, filling forms, copying data between applications — to automate repetitive, rule-based tasks. Unlike AI agents that reason and make decisions, RPA follows predefined scripts precisely. A Moroccan bank might use RPA to transfer data from email attachments into its core banking system. RPA is often a first step toward automation before adding AI capabilities like natural language understanding.
Robotics
Robotics is the interdisciplinary field combining AI, mechanical engineering, and electrical engineering to design, build, and operate robots that can perform physical tasks autonomously or semi-autonomously. Modern robotics integrates computer vision for perception, reinforcement learning for decision-making, and natural language processing for human-robot interaction. Applications range from manufacturing assembly lines and surgical robots to agricultural harvesting and warehouse logistics.
Rotary Positional Encoding (RoPE)
Rotary positional encoding (RoPE) is a technique for injecting positional information into transformer models by rotating the query and key vectors in attention layers, rather than adding position embeddings to inputs. RoPE encodes relative positions naturally, enabling models to generalize to sequence lengths not seen during training. It is used in LLaMA, Mistral, Qwen, and many other modern open-weight models, offering better length generalization than absolute positional encoding.
Scaling
Scaling in AI refers to the observation that model performance improves predictably as you increase compute, data, and parameters. This principle, sometimes called scaling laws, has driven the race to build ever-larger models. But scaling is not just about bigger models: it also means scaling operations, deploying AI to more users, more use cases, and more regions without breaking reliability or breaking the bank.
Self-Attention
Self-attention is the mechanism within transformer models that allows each token in a sequence to attend to every other token, computing relevance scores that determine how much each word should influence the representation of every other word. It is what lets a model understand that in «the cat sat on the mat», «it» refers to «the cat» and not «the mat». Self-attention replaced the sequential processing of RNNs with a parallelizable approach that scales efficiently.