01 · Introduction
What an LLM is
LLM stands for large language model. It is the program behind Claude, ChatGPT, Gemini and the other AI assistants.
An LLM has learned from a very large amount of text. From that it learned one skill extremely well: given some text, predict what text should come next. That one skill turns out to cover a lot: answering questions, writing, summarizing, translating, writing code and following instructions.
Three words you will meet everywhere:
- Model: one specific LLM, with a name and version, such as
claude-haiku-5-5. Bigger models are usually smarter and slower; smaller ones are cheaper and faster. - Prompt: the text you send to the model. Your question, plus any instructions and documents.
- Token: the small pieces of text a model reads and writes. You pay per token.
The model itself has no memory, no internet and no tools. Those are added around it by the app or the agent you use.
02 · When to use it
What LLMs are good and bad at
Good at:
- Turning rough notes into clean text, and long text into short summaries.
- Explaining something in simpler words, or in another language.
- Writing and fixing code, formulas and scripts.
- Reading a document you give it and answering questions about it.
- Sorting, tagging and extracting facts from many similar texts.
Bad at:
- Facts it did not see in your prompt. It will guess, and the guess can be fluent and wrong. This is called a hallucination.
- Exact arithmetic and counting. Give it a calculator tool instead.
- Anything that happened after its training cutoff, unless the app lets it search the web.
- Knowing when it is unsure. Always check anything with consequences.
03 · Setup
How to start using one
You do not install an LLM. You reach it in one of three ways:
- A chat app (Claude, ChatGPT): sign up, type, done. Best for learning what the model can do.
- An agent or tool that uses a model (Claude Code, an automation in n8n): you pick a model from a list and give it your API key.
- The API: your own program sends a prompt and receives the answer. You need an API key from the provider and a few lines of code.
Picking a model: start with the provider's mid-size model for everyday work, use the small one for simple bulk jobs like tagging or translation, and the largest one only for hard reasoning. The difference in price is often 10x or more.
04 · How to use it
Day-to-day use
- Say who the answer is for and what shape you want: "a 5-line summary for my manager, in plain words".
- Put the material in the prompt. The model cannot read your mind or your files unless you paste or attach them.
- Give one example of a good answer when the format matters.
- Ask it to show its reasoning or quote the source line when you need to check the result.
- Iterate. A second message that says "shorter, and drop the third point" is normal.
- Keep a note of prompts that work well and reuse them.
05 · Pitfalls
Pitfalls
- Confident mistakes. The tone is the same whether the answer is right or wrong. Verify names, numbers, dates and links.
- Privacy. Whatever you paste goes to the provider's servers. Check what your company allows before pasting customer data.
- Model updates. The same name can point to a newer model later; a prompt that worked may behave differently. Pin a dated version for anything automated.
- Long prompts cost more and can work worse. Send what matters, not everything you have.
- It is not a search engine. For current facts, use a tool that searches and then let the model summarize.
06 · Resources
Where to go next
Once the model makes sense, the next questions are what you give it (context), how it keeps things between chats (memory) and how it gets things done (agents).