LLM (Large Language Model)

A Large Language Model (LLM) is trained to predict and generate language sequences from large datasets. It can summarise, classify, extract or create text and code, but output is probabilistic and may contain fabricated facts, bias or unsafe instructions. The model does not automatically consult a current source or understand business accountability. Appropriate use defines permitted data, context, human review, evaluation, retention, suppliers and action limits. Adding retrieval or tools may improve usefulness and also increase impact. Tests should use representative cases and measure error, safety, latency and cost. Fluency is not evidence of truth; important output needs verifiable support and a fallback when the model fails. Connecting an LLM to actions requires stronger approval and monitoring than using it for drafts.