How an LLM gets useful
A model is first pretrained to predict the next piece of text across a huge collection of documents. That alone doesn't make a good assistant, so it is then trained further, for example with fine-tuning and RLHF (reinforcement learning from human feedback, where people rank answers and the model learns to prefer the better ones). Anthropic's glossary describes Claude this way.
What it means for you
An LLM writes a likely, useful continuation of what you give it. It doesn't look facts up unless the app gives it a search or RAG step, which is why answers need checking. What it can take in at once is limited by its context window, measured in tokens.
Model names change almost every week, so this glossary doesn't rank them. See What is an AI assistant, and which should I use?