LLM (large language model)

An LLM (Large Language Model) is an AI model with billions of parameters trained on large volumes of text to predict the next word. Useful skills emerge from that ability: drafting, summarising, translating, classifying, extracting data or writing code.

Under the AI Act, most LLMs are general-purpose AI models: their makers have documentation, copyright and transparency obligations, and the most powerful ones extra systemic-risk obligations. A company that uses them through a tool has the obligations that apply to its specific use.

Keep their limits in mind: they can hallucinate, reflect biases in their data and "know" nothing after their training unless connected to external sources.