Large Language Model
Aussi : LLM, Large Language Model, foundation model (language)
A Large Language Model is an AI system trained on vast text data to predict and generate language, enabling tasks like writing, summarizing, and answering questions.
What It Is
A Large Language Model (LLM) is a type of artificial intelligence trained on massive amounts of text to learn the statistical patterns of language. At its core, an LLM predicts the next token (a word or word fragment) given the preceding text. By doing this billions of times during training, the model builds an internal representation of grammar, facts, reasoning patterns, and writing styles.
Most modern LLMs are built on the transformer architecture, which uses a mechanism called attention to weigh how much each word in the input relates to every other word. Models are described by their number of parameters (the adjustable weights), often ranging from a few billion to hundreds of billions.
Why it matters
LLMs matter because they generalize across many tasks without needing task-specific training. The same model can draft an email, translate a sentence, extract data from a contract, or explain code. This flexibility lowers the cost of automating language-heavy work across data, marketing, finance, and AI teams.
How it is used in practice
- Content generation: drafting marketing copy, reports, or product descriptions.
- Summarization: condensing long documents, meetings, or research.
- Question answering: powering chatbots and internal knowledge assistants.
- Classification and extraction: tagging support tickets, pulling figures from invoices.
- Code assistance: generating and reviewing code.
Teams often improve results with prompt engineering (writing precise instructions), fine-tuning (further training on domain data), or retrieval-augmented generation (feeding the model relevant documents at query time to ground answers in trusted sources).
A concrete example
A finance analyst pastes a 40-page earnings report into an LLM-powered tool and asks: "Summarize revenue trends and list three risks mentioned." The model returns a short summary with bullet points in seconds. The analyst then verifies the figures against the source.
Limitations
LLMs can hallucinate (produce confident but false statements), reflect biases in training data, and lack real-time knowledge unless connected to external data. Human review remains essential for high-stakes decisions.