People say "the LLM" when they mean the model layer behind a chat product or API — distinct from the UI, agent harness, RAG stack, or fine-tune wrapped around it. Most current LLMs use transformer blocks and next-token objectives; prompting, retrieval, tools, or fine-tuning adapt those weights to downstream tasks. The acronym is used loosely: not every model marketed as an LLM is literally "large" by today's frontier scale, but the term stuck as the default label for generative text models.