How models work · L1 · 15 min
Tokens, not words
Learn in five minutes
- A model reads tokens, which may be parts of words.
- A space-split counter is a toy and is labeled as one.
- Token counts affect cost and context. Confirm a vendor price before you quote it.
- The 2017 transformer paper is the usual citation for the architecture family.
- Different models tokenize the same sentence differently.
Do not compare two products by a word count. Compare token counts from each tokenizer, and say which tokenizer you used.
Visual
One sentence, two counts
- Take the sentence 'do not split blindly'.
- Count spaces plus one and label it a toy.
- If you have a real tokenizer, record its name and its count beside the toy.
Quiz
Ecosystem
- Token
- The piece of text a given model reads or emits.
Sisters: Tokenizers, Transformers, Decoding, Model cards.
The same next-token idea sits under chat, voice, and tool calls.
Pin a model id when you leave the toy explanations.
Config
- Write the goal in one sentence.
- Name the input and the allowed output.
- Save the trace next to the result.
Troubleshooting
- The answer looks fluent and is wrong. Check whether a source passage actually contains the claim.
- The loop never stops. Set a step cap and stop when the tool result repeats.
Sources and go deeper
- Vaswani and co-authors, Attention Is All You Need, 2017 paper last checked 2026-10
- Brown and co-authors, Language Models are Few-Shot Learners, 2020 paper last checked 2026-10
- Attention Is All You Need — Wikipedia wikipedia last checked 2026-10
- Large language model — Wikipedia wikipedia last checked 2026-10