Language Model

What is a Language Model?

A Language Model is a neural network trained to process and generate text. It uses the same underlying architecture for very different tasks—translation, summarization, programming—without requiring separate training for each task. Claude, ChatGPT, and Gemini are well-known examples of language models.

How it works

A language model processes text as sequences of numbers. A tokenizer splits text into tokens, and the neural network computes a probability distribution over possible next tokens for each token. Modern language models are based on the Transformer architecture with Self-Attention, which relates tokens to one another instead of processing them only in a fixed sequence.

Why Language Models shape the public perception of AI

Language models are the first AI application that can be used without technical expertise—a text box is enough. Earlier AI systems mostly operated invisibly in the background, for example as spam filters or credit-scoring systems. Language models are visible, directly accessible, and produce answers that appear human. They now shape the public perception of artificial intelligence as a whole.

A language model is not a synonym for Artificial Intelligence but one specific area within it. Likewise, a language model does not “understand” text in the human sense. It computes probabilities for token continuations without any independent concept of truth or meaning beyond the current context.

Why Language Models have practical implications

• Versatility: The same model architecture handles tasks ranging from translation to programming, controlled entirely by the prompt.
• Not a knowledge database: A language model does not retrieve stored facts like a database. Instead, it computes plausible continuations, which also explains why hallucinations can occur.
• Resource requirements: The size and practical operation of a language model depend directly on the number of parameters, the context window, and quantization.

Typical example

The same model architecture that summarizes a document can, without additional training, also write code or translate text. The task is determined by the prompt rather than by a different version of the model.

Further reading

Fundamentals
→ How a Language Model Works
→ Artificial Intelligence as an Umbrella Term

Related terms
→ Artificial Intelligence
→ Neural Network