Skip to content
AIOpenAlt AI Labs

Prompting 101: five real tasks, before and after

A free starter challenge. Learn how language models work and how to write prompts that get useful answers, then prove it on five real tasks from your life, each with a before and an after.

  1. 1Learn
  2. 2Do
  3. 3Submit
  4. 4Results
Category
AI
Lesson
20 min
Effort
~2h
Prize
Free
Closes in
 

Learn

20 min lesson

What you'll learn

How a language model turns your words into an answer, and how to write prompts that get useful results the first time. You will practise on five real tasks from your own life and compare a lazy prompt with a careful one.

How it works, in one paragraph

A language model predicts the next piece of text, one piece at a time, based on everything you gave it. It does not know who you are, what you already tried or what "good" means to you unless you say so. Vague in, vague out. The 3Blue1Brown video shows this in eight minutes; watch it first.

The five parts of a good prompt

  1. Role or situation. "I am a first-year student in Osh applying for a summer internship."
  2. The task. One clear verb: write, explain, compare, check, plan.
  3. Context. Paste what the model needs: the job post, your notes, the text to fix.
  4. Constraints. Length, language, tone, what to avoid. "Under 150 words, in Russian, polite but not stiff."
  5. Format. "A table with three columns", "five bullet points", "an email with a subject line".

You will not need all five every time. But when an answer disappoints you, one of them is usually missing.

Iterate instead of starting over

Treat the first answer as a draft. Say what is wrong: "too formal", "you invented a number, use only the data I gave you", "shorter". Ask the model to ask you questions first when the task is big: "Before you write, ask me three questions you need answered."

Check what comes back

Models sound confident even when wrong. Names, dates, numbers, quotes and laws need checking against a real source. Never paste passwords, other people's personal data or anything secret into a chat.

Your five tasks

Pick five real tasks you actually need done. Good ones: a cover letter for a real job post, a study plan, an explanation of a hard topic, a polite message to a teacher, a summary of a long article. For each one, write it down in your doc like this:

  • Task: what you needed
  • Before: your first, lazy prompt and a short piece of the answer
  • After: your improved prompt and a short piece of the better answer
  • What changed: one or two sentences on which part of the prompt made the difference

Use any model you like: ChatGPT, Claude, Gemini, DeepSeek.

Before you submit

Put everything in one Google Doc or Notion page, share it with "Anyone with the link can view", and open it in a private window to check.

Finish the lesson to unlock submitting.

Sign in to start learning

Do

about 2h of work

The task

Most people use AI chat tools the way they use a search box, and get average answers. The people who get great answers give the model a situation, a clear task, the context it needs, constraints and a format, and they iterate. In this challenge you pick five real tasks you actually need done, such as a cover letter, a study plan or a hard explanation, and for each one you show your first lazy prompt, your improved prompt, a piece of each answer, and what made the difference.

Deliverables

  1. 01One shared Google Doc or Notion page with all five tasks
  2. 02For each task: the task, the before prompt and answer excerpt, the after prompt and answer excerpt
  3. 03For each task: one or two sentences on what changed and why it worked

Acceptance criteria

Judged against exactly these

  • There are five different, real tasks, not five versions of the same one
  • Each after prompt is clearly better than its before prompt, using at least three of: situation, task, context, constraints, format
  • Each pair includes a short excerpt of both answers, so the difference is visible
  • Each task explains in your own words which change made the difference
  • At least one task shows a fact or number from the model being checked against a real source
  • No passwords, secrets or other people's personal data appear in the prompts

Rules

  • Complete the lesson first. The submit form unlocks once you mark it done.
  • Submit a link to your work: a public GitHub repo or PR for code challenges; YouTube, Google Drive or Docs, Figma, Notion or a live site for everything else.
  • AI tools are allowed and encouraged. Plagiarised or copied work is disqualified.
  • The work must be made during the challenge window. For code, the commit history must show it.
  • An AI reviewer pre-scores every entry against the acceptance criteria. A human makes the final decision.
  • On payout, Alt AI Labs receives a licence to use the winning work (MIT for code). Non-winning entries stay yours.
  • Do not include secrets, API keys or anyone's personal data.
promptingllmwritingproblem-solving

Submit

Sign in and complete the lesson to submit. Entry is free.

Sign in

Results

  1. 01

    AI pre-score

    AI reviews your write-up against the acceptance criteria. It cannot open your link, so explain your work clearly.

  2. 02

    Human decision

    A human reviews the work itself and makes the final call. The AI score is advisory.

  3. 03

    Win

    There is no cash prize. Winning work is shown to the organization behind the challenge.

Posted by

Alt AI Labs