What is an LLM

A large language model (LLM) does one thing: given some text, it predicts the next chunk of text. It was trained by reading an enormous amount of writing - books, code, articles, forums - and learning the patterns in how words follow one another. When you send it a question, it isn't looking anything up. It's generating the most plausible continuation of what you wrote, one chunk at a time.

That single idea explains almost everything about how these models behave, so it's worth holding onto: an LLM produces text that sounds right, which is not the same as text that is right.

What follows from "predict the next chunk"

A few consequences fall straight out of how the model works, and they shape how you use it:

• It has a knowledge cutoff. The model only knows what was in its training data. Ask about something that happened after it was trained, or about a private repo it never saw, and it has nothing to draw on.
• It doesn't fetch live data on its own. No web search, no database, no clock, unless a tool is wired up to give it those. On its own it only has the text you send plus what it learned in training.
• It isn't deterministic. Send the exact same prompt twice and you can get two different answers. There's deliberate randomness in how it picks each next chunk.
• It can be confidently wrong. Because it generates plausible text, a wrong answer looks just as fluent and self-assured as a right one.

What it's genuinely good at

None of that makes the model weak - it makes it a specific kind of tool. LLMs are strong at working with language: drafting and rewriting, summarizing, explaining, translating, turning messy notes into structure, writing and reviewing code, and answering questions about text you hand it.

They're weak wherever fluent-sounding text isn't good enough on its own: exact arithmetic, counting, and any fact you can't afford to have invented. For those, you give the model the real data to work from or you check its output. The rest of this skill is largely about doing exactly that well.

Next: the "chunks" the model reads and writes have a name - tokens - and nearly everything about a request is measured in them.

Your lab setup

This VM comes preconfigured with your AI keys - just type claude to get started. Your key and URL are already set as $OPENAI_BASE_URL and $OPENAI_API_KEY.