Large language model

Germany's Federal Office for Information Security (BSI) explains in a study on generative AI models (in German) that large language models generate text from correlations learned in training. They use probability distributions to predict which character, word or sequence of words might come next. Article 3(63) of the EU AI Act defines a related term, the general-purpose AI model AI Artificial intelligence (AI) covers methods that let a computer perform tasks which would require intelligence if a person did them. Read in the glossary → . Among other criteria, such a model is not tied to one purpose, handles a wide range of different tasks competently and can be built into many other systems. Recital 99 cites large generative AI models as a typical example.

A model without access to real-time data has no information about current events, the BSI notes. A shop's current stock levels are not in the model, so the application has to supply them, for instance through RAG RAG Retrieval-augmented generation (RAG) is a method that lets a generative AI model draw on documents held in a knowledge base and use them to enrich its output. Read in the glossary → . We describe GEO GEO GEO, or generative engine optimisation, means making content more likely to appear in the answers given by generative AI search. Read in the glossary → as building content so that language models understand, cite and recommend it.

We use models from external providers and neither run nor train our own. We choose the model for each step: in one project, a small model handles simple tasks, a mid-sized one long prompts Prompt A prompt is the input a user gives a generative AI system, which then responds with an output. Read in the glossary → and the largest demanding reasoning. We are not tied to one provider, and the model behind an agent can be swapped. Local models are possible; we judge case by case whether one is reliable enough.