powstik
Prove
LiveProof-of-Work · Oct–Nov: every verified proof counts. 35 days leftAll outcomes
Skills

Learn Ethics, Safety & Alignment (AI Engineering Phase 18)

5 weeks · 2 milestones

Phase 18 of AI Engineering from Scratch: 30 free lessons, then one project of your own that is run, evaluated and documented. Part of the Varsity Curriculum: AI Engineering.

Milestone map

Milestone map

2 milestones

Do the phase's lessons in public

1–2 weeks (about 10 hrs/week, est)

Work through the 30 lessons of Phase 18: Ethics, Safety & Alignment from AI Engineering from Scratch (free) and keep the ethics, safety and alignment lessons' notebooks (reward hacking, red-teaming, prompt injection, bias and fairness, privacy, model cards) in one public repo as you go. About 17 hours (est).

Proof required

Link your public repo with the ethics, safety and alignment lessons' notebooks (reward hacking, red-teaming, prompt injection, bias and fairness, privacy, model cards), and show your self-check (quizzes passed, notebooks run, tests passing where they exist).

What gets checked

  • Each notebook runs and its quiz is passed.
  • Red-team work is done only on your own app or one you are allowed to test.
  • Findings are written with evidence.

Common mistakes

  • Copying the repo's solutions: verifiers read your history and ask about your code.
  • Skipping the self-check, then asking for review too early.

Powstik Guide

A public repo with the ethics, safety and alignment lessons' notebooks (reward hacking, red-teaming, prompt injection, bias and fairness, privacy, model cards).

Steps

  1. Make a public repo for this phase.
  2. Do one lesson at a time and commit after each.
  3. Self-check: Run each lesson's code; it must work on your machine as the lesson shows. Run each lesson's notebook top to bottom; its outputs must match the lesson. Each lesson has a quiz (quiz.json, also on the course site): pass it.
  4. Write a short README: what is here and how to run it.
  5. Ask three people who know the topic to check it.

Template

README
Phase 18: Ethics, Safety & Alignment (AI Engineering from Scratch)
What is here:
How to run it:
Self-check result:
What was hardest:

What gets sent back

  • One big upload with no history.
  • Self-check not shown.
  • Code that does not run from the README.

Resources

Foundationstart here

Depthgo deeper

What a verifier looks for

  • Each notebook runs and its quiz is passed.
  • Red-team work is done only on your own app or one you are allowed to test.
  • Findings are written with evidence.
  • Ask about one choice in the code; the answer should be clear.

You'll sign in first, then come straight back here.

Red-team an AI app and fix it

2–4 weeks (6–10 hrs/week)

Use what you learned. Make a red-team and evaluation report on an AI app (yours, or an open-source one you may test): the attacks tried (prompt injection, jailbreaks, bias checks), what failed, the fixes shipped, and a model or system card. It must run, be evaluated and be documented. Include a short written evaluation: what you tested, what failed and what you changed.

Proof required

Link a red-team and evaluation report on an AI app (yours, or an open-source one you may test): the attacks tried (prompt injection, jailbreaks, bias checks), what failed, the fixes shipped, and a model or system card, with a README that says how to run it and what the evaluation found. Include a short written evaluation: what you tested, what failed and what you changed.

What gets checked

  • Only an app you may test was attacked.
  • Each finding has evidence and a fix, re-tested after.
  • The written evaluation covers what was tested, what failed and what changed.

Common mistakes

  • A demo with no evaluation.
  • Numbers without the data or code that produced them.

Powstik Guide

A red-team and evaluation report on an AI app (yours, or an open-source one you may test): the attacks tried (prompt injection, jailbreaks, bias checks), what failed, the fixes shipped, and a model or system card.

Steps

  1. Write one line: what it does and for whom.
  2. Build the smallest version that works.
  3. Write the evaluation first, then improve against it.
  4. Write the README with the results.
  5. Ask three people to check it.

Template

README
What it does:
How to run it:
How it was evaluated:
Results:
What was tested, what failed, what changed:
Limits:

What gets sent back

  • Does not run from the README.
  • No evaluation.
  • Claims more than the evaluation shows.

Resources

Foundationstart here

What a verifier looks for

  • Only an app you may test was attacked.
  • Each finding has evidence and a fix, re-tested after.
  • The written evaluation covers what was tested, what failed and what changed.
  • Run it, or watch it run: it must work, be evaluated and be documented.
  • Check that the work is the person's own.

You'll sign in first, then come straight back here.

Part of