Large Language Model

A large language model (LLM) is a neural network trained on very large text corpora to predict the next token in a sequence, a simple objective that turns out to support generation, translation, summarization, question answering, and code writing. Nearly all current models use the transformer architecture, whose attention mechanism lets the model weigh relationships between distant parts of the input. Training happens in stages: pretraining on broad text, then instruction tuning and alignment steps such as RLHF that turn a raw predictor into a usable assistant. Capability generally rises with model size, data volume, and training compute, and more recently with computation spent at inference time on reasoning. OpenAI's GPT models, Anthropic's Claude, Google's Gemini, and open weight families such as Llama, Qwen, and Mistral are the widely used examples, reached through APIs, chat interfaces, or self-hosting. The persistent limitation is that an LLM predicts plausible text rather than retrieving verified facts, so it can produce confident and wrong statements, a failure mode called hallucination. Retrieval-augmented generation, tool use, and citation requirements are the standard mitigations, and a model's knowledge is fixed at its training cutoff unless fresh context is supplied.