A large language model predicts likely sequences of text, and at sufficient scale that ability supports summarizing, drafting, answering questions, extracting information, writing code, and following multi-step instructions. LLMs are offered by several providers, and some can be run on private infrastructure.
LLMs are powerful but not authoritative. They can produce fluent text that is wrong, and their output depends heavily on the instructions and context they receive. Reliable business use puts them inside a system with retrieval of trusted information, clear permissions, evaluation, and human review where the stakes require it.