Stage 1 of 6 · 6 lessons · ~2 weeks

What AI data work is and why it exists

Understand why AI models need people: labelled examples for traditional models, written demonstrations and preference choices for chatbots, and human testing for quality and safety. Then see who pays for this work and what the 2026 demand data actually says.

Start: Labelled data and why models need it

Lessons· one lesson for each topic, in this order

Do this

Explain demonstrations, preferences and RLHF to a friend in two minutes.
Do five side-by-side comparisons on Arena, writing a justification before each vote.
Open a public preference dataset on Hugging Face and read 20 pairs.
List three data vendors and the platform each uses to recruit.
Note the Upwork and LinkedIn 2026 demand facts with their sources and dates.
Write down which of your subjects or languages you could judge at expert level.

Start with· follow one of these

CoursePaid
Grokking Modern AI Fundamentals
Design Gurus · Beginner · $72 one-time
A structured course on how modern AI models are built and trained, for the full picture behind your tasks
VideoFree
Deep Dive into Large Language Models like ChatGPT
Andrej Karpathy · Intermediate
Free: Karpathy walks through pre-training, fine-tuning and RLHF, including the role of human labellers

Practise

Arena (formerly LMArena)
Compare two anonymous answers and vote: the core motion of preference work, free
Hugging Face Hub
Open public labelled and preference datasets and read real examples of what this work produces

Keep handy

Upwork
Search job posts for data annotation and AI training to see what clients ask for today
LinkedIn Jobs
Search "data annotator" and "AI trainer" in your country to see which companies hire locally

Also worth your time

Illustrating Reinforcement Learning from Human Feedback (RLHF) (opens in a new tab)Read
A clear, illustrated walk through demonstrations, reward models and tuning
Reinforcement Learning From Human Feedback (DeepLearning.AI)Go deeper
A free 90-minute course on the preference and prompt datasets RLHF uses and how tuned models are compared

Before moving on: A one-page note in your own words: what labels, demonstrations, preference data, evals and red teaming are; the data supply chain from lab to platform; the two demand facts above with their sources; and which kind of task you want to do first.

Tauseef's advice

Spend a few hours understanding where your work goes before you apply anywhere. People who know why a guideline exists follow it better, and they can explain their work in applications and later in data or ML interviews. It also makes the scams easier to see.