LLMs are trained in two stages: unsupervised pre-training on internet-scale text, then fine-tuning and RLHF (reinforcement learning from human feedback) that shape them into helpful, safer assistants.
They're used for writing, coding, summarisation, translation, classification, and — through function calling and RAG — as the reasoning layer of larger applications.
LLMs hallucinate: they can produce fluent, confident, wrong answers. Ground them with retrieval, verification, and human review whenever accuracy matters.