Prompt-and-response intuition

What it is

A prompt is what you ask an AI — your instructions. The model stays the same; the prompt is the steering wheel. This toy bot knows one fixed set of facts and never learns anything new, yet by changing the topic, the style, and the length of your prompt you get wildly different answers from the very same bot.

Go deeper: good prompting (sometimes called prompt engineering) usually means three things — say what you want (the topic), say how you want it (the style or role), and say how much (the length or detail). Leave those out and the model has to guess, so it gives a vague or generic answer. The clearer the instructions, the better the result.

Why care

Most people’s first frustration with AI is “it didn’t give me what I wanted.” Often the fix isn’t a smarter model — it’s a clearer prompt. Learning to ask well is now a real skill, and it starts with seeing that the answer you get is shaped by the question you ask.

The idea, intuitively

Imagine asking a friend who knows a lot about volcanoes. “Tell me stuff” gets a shrug. “Explain volcanoes to my little brother in one sentence” gets something useful. Same friend, same knowledge — the difference is entirely in how you asked.

Peek at the data first

The bot’s knowledge is a tiny fixed list, and your prompt is built only from menu choices — you never type free text (safety by design). The same facts can come back many ways.

Try it

Pick a Topic, a Style, and a Length and read the bot’s reply. Keep the topic fixed and switch the style — same facts, new answer. Then tick Make the prompt vague to see what a fuzzy prompt gets you.

Where it shows up

• Chatbots. Telling them the role, format, and length you want makes answers far more useful.
• Image generators. The words in your prompt decide the style, subject, and mood of the picture.
• Search & assistants. A specific request beats a vague one almost every time.

Where it came from

Prompting became a topic when large language models like GPT-3 (2020) showed they could do new tasks just from a worded instruction — no retraining needed. Researchers found that how you phrase the request (for example, asking a model to “think step by step,” Wei et al., 2022) can dramatically change the answer, and “prompt engineering” was born.

Try it in code

Spectra doesn’t chat, but the “same model, different instructions” idea shows up when you ask the word-babbler to generate in different amounts:

data = load "sayings"
model = make_model "markov"
train_model model, on: data, using: "text"

generate model, count: 1
generate model, count: 5

Open it in the Studio ▶

Check your understanding

• If the answer changed, did the bot’s knowledge change? What changed instead?
• What three things make a prompt clear?
• Why does a vague prompt usually get a vague answer?