Sefism AI Data and Training Work Cheat Sheet

What this cheat sheet is

The Sefism AI data and training work cheat sheet is a revision reference for paid AI data work: labelling data, or writing, rating and fixing model answers. It is written for students and graduates doing this work or about to start, and it follows the same order as the AI data and training work roadmap.

Each edition has 9 chapters: 66 concepts for data annotation and 64 for AI trainers. Each concept is written the same way: what it is in plain words, why it matters, and the mistake people usually make with it. Each chapter ends with up to three library resources, and a glossary covers the smaller terms. Platform facts are as of October 2026.

Chapter by chapter

1. What AI data work is

Labelled data, demonstrations (SFT data), preference data and RLHF, reward models, evals, red teaming, the AI data supply chain and the 2026 demand evidence. Knowing where your work goes makes guidelines easier to follow and scams easier to see.

2. The honest picture

Five concepts: pay that varies, your effective hourly rate, project-based work, independent contractor status and sensitive content work.

3. Scam red flags

Seven scams aimed at people looking for this work: paying to start, fake recruiters on chat apps, top-up task scams, overpayment and cheque tricks, ID and bank detail harvesting, bought, rented or shared accounts, and look-alike sites and names. Real platforms are free to join, recruit on their own sites and pay through named services.

4. Core skills every test checks

A guideline-reading checklist, edge cases, gold tasks, inter-annotator agreement, Cohen's kappa, accuracy before speed, and the rule against AI tools in tasks unless a project allows them.

5. Your path's craft

One chapter per edition. Annotation covers bounding boxes, polygons and segmentation masks, keypoints, video object tracking, text classification, named entity recognition, audio transcription, Label Studio, CVAT and export formats. The AI trainer chapter covers ideal responses, rubric rating, Likert scales, rubric examples, side-by-side comparison, written justifications, fact-checking model answers, hard and adversarial prompts and reviewing AI-generated code.

6. Getting accepted

Qualification assessments, country availability, identity verification, an honest and specific profile and applying to several platforms, with a card on annotation clients on Upwork in the annotation edition.

7. Platform comparison

Seven platforms as their own pages describe them in October 2026: Outlier (Scale AI), DataAnnotation, TELUS Digital AI Community, CrowdGen (Appen), Mindrift (Toloka), Alignerr (Labelbox) and Turing. It deliberately gives no pay rates, because they vary by project and location and change often.

8. Track record and getting paid

Quality scores, payout methods, where PayPal cannot receive, a payout log, taxes as a contractor in general terms, PSEB registration in Pakistan, the freelancer ID in Bangladesh, and confidentiality terms.

9. Moving up

QA reviewer, team lead and project roles, specialist and expert projects, evals work, and routes into data analytics, machine learning and AI engineering, with the public proof each one needs.

Rules of thumb

The sheet closes with nine tips per edition, among them: never pay anyone to get, keep or qualify for AI data work; check a platform's eligibility page and payment methods for your country before you take its test; read the whole guideline before the first item; accuracy first, because speed comes from practice; keep your account, ID, location and payment method in your own name; log every payout and the hours behind it from day one; and treat any single project as temporary. Annotators get "tight boxes, consistent classes, every attribute filled"; AI trainers get "run the code and check the source before you rate an answer".

How to use it with the roadmap

• After each stage, read the matching chapter and see whether you can explain every concept without looking.
• Before any qualification test, reread chapters 4 and 5. Most failed tests are guideline mistakes, not knowledge gaps.
• Before you share documents or accept a task offer, run through chapter 3.
• Once payouts start, use chapter 8 to set up your records and check your country's rules.

How it works on Sefism

Everything on Sefism is free to learn from. The public cheat sheet page shows every chapter title, with the first chapters open to read. Keeping the sheet uses credits and opens every chapter in your member area; if you add the AI data and training work roadmap to your dashboard, its cheat sheet comes with it, and the first roadmap you add is free. New members get starting credits, and if you have a question the sheet does not answer, you can send it to Tauseef.

Related on Sefism

• Label Studio and CVAT for the annotation chapter.
• Google Search Quality Rater Guidelines for practising long guidelines.
• Reinforcement learning from human feedback and Grokking Modern AI Fundamentals for chapter 1.
• FTC: job scams for chapter 3, and the PSEB freelance programme for chapter 8.
• Related roadmaps: Freelancing and Remote jobs.

FAQ

Is this a substitute for the roadmap? No. The roadmap is the plan, with hours, milestones and checklists; the sheet is what you revise from once you have done the work.

What is inter-annotator agreement? A measure of how often two people label the same item the same way. Platforms use it, along with gold tasks, to check your consistency. Chapter 4 explains Cohen's kappa in plain terms.

Why does the platform comparison show no pay rates? Because rates vary by project and location and change often. The sheet sends you to each platform's official FAQ instead.

Can I use ChatGPT or another AI tool to do tasks faster? Not unless the project explicitly allows it. Most platforms forbid it and check for it, and it can cost you your account.