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
Underfitting
Underfitting occurs when a model is too simple to capture the underlying patterns in the data, performing poorly on both training and test data. Unlike overfitting (which memorizes), underfitting fails to learn at all — like a student who doesn't study enough and scores poorly on every exam. Causes include an overly simple model, insufficient training time, or inadequate features. Remedies include using a more complex model, training longer, adding features, or reducing regularization.
Unsupervised Learning
Unsupervised learning is a branch of machine learning where the model finds patterns and structure in data without labeled examples or human guidance. The most common tasks are clustering (grouping similar items) and dimensionality reduction (compressing features). It is useful when labels are expensive or impossible to obtain: a retailer discovers natural customer segments, a researcher identifies gene groups in biological data, and a security system detects unusual network behavior.
Vector Database
A vector database is specialized storage optimized for indexing and retrieving high-dimensional vectors (embeddings) efficiently. Unlike traditional databases that match exact values, vector databases find the nearest neighbors in embedding space, enabling semantic search, similarity matching, and RAG systems. Popular options include Pinecone, Weaviate, Qdrant, and Milvus. They are the backbone of AI applications that need to search by meaning rather than keywords.
Vector Search
Vector search is a technique for retrieving semantically similar items by comparing their vector embeddings rather than exact keyword matches. It powers recommendation systems, semantic search engines, and retrieval-augmented generation pipelines. Libraries like FAISS, Pinecone, and Weaviate enable efficient nearest-neighbor search across millions of high-dimensional vectors.
Vibe Coding
Vibe coding is an emerging practice where developers describe what they want in natural language and let AI coding agents generate the implementation, focusing on intent rather than syntax. The developer reviews, guides, and corrects the output. Tools like Replit Agent, Cursor, and GitHub Copilot enable this workflow. It dramatically speeds up prototyping and lets non-developers build functional tools, though production-quality code still needs expert review.
Vision-Language Model (VLM)
A vision-language model is a multimodal AI system that processes and reasons about both images and text simultaneously, understanding the relationship between visual and textual content. Examples include GPT-4V, Gemini, LLaVA, and Claude 3 Vision. VLMs can answer questions about images, generate text descriptions of visual scenes, perform visual question answering, and follow complex multimodal instructions. They are transforming how businesses interact with visual data — from analyzing product photos to reading charts and documents.
Weaviate
Weaviate is an open-source vector database designed for AI-native applications. It stores and indexes vector embeddings alongside their original data, enabling semantic search, recommendation systems, and retrieval-augmented generation (RAG). Weaviate supports hybrid search (combining vector and keyword search), automatic vectorization through integrated ML modules, and scales horizontally. A Moroccan startup building an AI-powered product search could use Weaviate to let customers find items by meaning rather than exact keywords.
Workflow Automation
Workflow automation connects multiple software tools and AI components into a seamless sequence that executes business processes without manual intervention. Traditional workflow automation follows fixed rules (Zapier, Make); AI-enhanced workflow automation adds language understanding, decision-making, and adaptation. A Moroccan logistics company might automate order processing: receive an order via email, extract details with AI, check inventory in the ERP, and trigger shipping — all without a human touching the process until an exception occurs.
XGBoost
XGBoost (Extreme Gradient Boosting) is a highly efficient machine learning algorithm based on ensemble learning, specifically gradient-boosted decision trees. It builds trees sequentially, with each new tree correcting errors from the previous ones. XGBoost consistently wins Kaggle competitions and is a go-to algorithm for tabular data tasks like fraud detection, customer churn prediction, and credit scoring. It is fast, handles missing values natively, and includes built-in regularization to prevent overfitting.