💭 in your head · nothing to type
Picture walking into a workshop where everyone has known each other for years. On the wall is one toolbox with six instruments: token, context, temperature, model, prompt, and agent.
Which instrument would you pick up first to follow the conversation?
💭 in your head · nothing to type
Return to the workshop: six instruments share one toolbox, but each serves a different operation. You do not need to rebuild an engine to drive; you need to recognize the dashboard lights.
Which instrument will you learn to recognize first?
"AI", "neural network", "machine learning" and "LLM" are often used as synonyms. In fact they sit inside one another, like a nesting doll.
<aside> 🧭
Interactive block — on the site: open →
</aside>
AI is the biggest umbrella: any machine doing something "smart", from a chess program to a sat-nav. Inside it is machine learning: instead of writing the rules, you show the program lots of examples and it works the rules out itself. Inside that is deep learning: the same idea, but on many-layered neural networks. And right in the middle are LLMs, large language models trained on huge amounts of text. Claude and ChatGPT live here.
Why it matters: when the news says "AI has learned to…", it could mean anything from the four layers. Everything you'll use in this course is the smallest doll.
The idea. A model reads text not letter by letter or word by word, but in pieces called tokens. A token can be a whole short word, part of a long word, or a punctuation mark. Everything in the LLM world is measured in tokens: request length, memory size, price.
Analogy. A construction set: before text goes into the model, it's taken apart into bricks. A short common word is one brick, a rare long one is several.
See it yourself. Open any online tokenizer (for example, platform.openai.com/tokenizer) and paste one sentence in English, then the same sentence in another language you know. The non-English version usually takes more tokens: it gets chopped finer. The practical takeaway: long non-English text eats through limits faster.