Chapter 01 · Section I · 8 min read
1.1 What a prompt actually is
A prompt is just text — but text that the model treats as a complete instruction, not a conversation partner. Getting the mental model right before you learn any tricks.
A prompt is the block of text you send to a language model to get a response back. That’s the whole definition. Everything else in this course is craft — how to write that text so the response is closer to what you actually wanted.
The trouble is that the word prompt sounds conversational, so people write prompts the way they’d write a chat message to a friend: half a thought, some assumed context, an implicit ask. Models are not friends. They cannot ask you a clarifying question mid-sentence, they do not remember what you said last week, and they have no idea what you were looking at on your screen when you typed.
The one useful metaphor
Think of the prompt as a note to a very literal, very well-read intern on their first day. They will:
- Do exactly what the note says.
- Fill in gaps with the most statistically plausible guess.
- Not question the premise, even if the premise is broken.
- Forget the entire conversation the moment they finish the task.
Every skill in this course is a way to write a note that leaves fewer gaps and gives the intern less room to guess wrong.
What a prompt is not
- Not a conversation. Each call to a model is (usually) fresh. If you don’t include the context, it isn’t there.
- Not a query. Google matches your words to indexed pages. A model generates new text based on patterns from training. Asking a model “what is the capital of Bhutan” and asking it “write a limerick about the capital of Bhutan” are the same kind of operation to the model — both are text generation.
- Not a database lookup. The model will confidently produce plausible-looking output for questions it has no real answer to. This is the source of most hallucinations.
- Not deterministic. The same prompt sent twice can produce different responses. Most of the time this doesn’t matter; sometimes it matters a lot.
A first side-by-side
A weak prompt:
Write about the earthquake.A stronger prompt:
Write a 150-word news brief for a Nepali high-school newspaper about the
2015 Gorkha earthquake. Cover: when it happened, magnitude, districts most
affected, and one specific way the country responded. Plain English, no
statistics the reader can't verify.The second prompt still fits in a few lines, but it names the audience, the length, the shape, and the four things the reader must get out of it. There is almost no room for the model to guess wrong.
Check your understanding
Quick check
—