Language models are AI systems trained to predict and generate text. They power chatbots, autocomplete, translation, summarization, and code generation. Modern language models like GPT, Claude, and Gemini are trained on trillions of words and have become surprisingly capable at tasks they were never explicitly programmed to do.

A language model's core job is conceptually simple: given a sequence of words, predict the next word. Repeat this prediction one word at a time and you get text generation. The simplicity of this objective hides extraordinary complexity in execution. Modern language models are deep neural networks based on the transformer architecture introduced in 2017. They're trained on massive text corpora — books, websites, code repositories, academic papers — and learn statistical patterns that capture grammar, factual associations, reasoning patterns, and writing styles. Pretraining on raw text produces a base model with broad capabilities but no specific behavior. Fine-tuning on instruction-following examples and reinforcement learning from human feedback (RLHF) shape the model into a usable assistant that follows directions, refuses harmful requests, and matches a desired conversational style. Today's frontier language models — GPT, Claude, Gemini, and their open-source counterparts like Llama and Mistral — can write coherent essays, debug code, translate between dozens of languages, summarize long documents, and reason through multi-step problems. They are not databases (they hallucinate facts), not reasoning engines in any deep sense (they pattern-match plausible reasoning), and not understanding minds (they process tokens, not meaning). But within those limitations, they have become the most generally useful AI systems ever built and the foundation for most modern AI applications.

What Are Language Models?