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
Text Generation
Text generation is the task of automatically producing human-like text given a prompt or context. Modern text generation uses large language models that predict the most likely next token iteratively. Applications include email drafting, content creation, code writing, summarization, translation, and chatbot responses. Quality depends heavily on prompt design, model choice, and parameters like temperature and top-p sampling.
Text-to-Speech (TTS)
Text-to-speech converts written text into spoken audio output using AI models. Modern TTS systems produce natural-sounding, expressive speech with appropriate prosody, emotion, and pacing. They power virtual assistants (Siri, Alexa), accessibility tools for visually impaired users, automated phone systems, audiobook narration, and content localization. Neural TTS models like those from ElevenLabs and OpenAI can clone specific voices from short audio samples, enabling personalized experiences.
Time-Series Forecasting
Time-series forecasting predicts future values based on historically observed sequential data points — stock prices, website traffic, energy consumption, or weather patterns. AI approaches include traditional statistical methods (ARIMA), classical ML (random forests on engineered features), and deep learning (LSTMs, transformers, temporal fusion transformers). A Moroccan retailer could forecast demand for each product in each store to optimize inventory, reducing both stockouts and overstock waste.
Token
A token is the smallest unit of text that a language model processes. For English, a token is roughly 4 characters or ¾ of a word; for Arabic and French, token boundaries vary. Models use token counts to manage their context window and to calculate costs. Understanding tokens matters for prompt engineering: longer prompts consume more tokens, and every model has a maximum token limit per request.
Tokenization
Tokenization is the process of breaking raw text into tokens — the units a language model actually reads. Different tokenizers handle words, subwords, or characters depending on the model. Word-level tokenization splits on spaces; subword tokenizers like BPE merge common character sequences to handle rare words. Proper tokenization is critical for multilingual models that must handle Arabic, French, and English in the same system.
Tool Use
Tool use is the ability of an AI agent to invoke external software tools — web browsers, APIs, databases, calculators, file systems — to accomplish tasks beyond what the language model can do with text alone. When a model calls a search API, writes to a database, or sends an email, it is using tools. Tool use transforms a language model from a text generator into an active participant in business workflows, bridging the gap between understanding and action.
Training Data
Training data is the dataset used to teach a machine learning model by example, allowing it to learn patterns, relationships, and rules. The quality, quantity, and representativeness of training data directly determine model performance. Biased or insufficient training data produces biased or weak models. Training data is typically split into training, validation, and test sets to evaluate generalization properly.
Transfer Learning
Transfer learning takes a model trained on one task and repurposes it for a different but related task, saving enormous amounts of data and compute. A model pre-trained on millions of general images can be fine-tuned with just a few hundred medical scans to detect diseases. This is why modern AI development starts with pre-trained foundation models rather than training from scratch, dramatically lowering the barrier to entry.
Transformer
The transformer is the neural network architecture that revolutionized NLP and underpins all modern large language models. Its key innovation is the self-attention mechanism, which processes all tokens in parallel rather than sequentially, enabling massive speedups and much better long-range context handling. GPT, BERT, Claude, Gemini — they are all transformer-based. Transformers also power vision models, speech systems, and protein-folding research.