AI Instructor – Preference ranking and RLHF (Remote, Western Europe and the Nordics)

What the work is for

YPAI Learn prepares people for AI evaluation work in live classes. Your classes take learners from picking a response by feel to pairwise preferences they can explain, and teach them to check that their reasons hold across many comparisons and to see how those preferences shape a model's behaviour.

The work

• Teach the live class “Choose the better of two AI responses and justify the preference”, 90 minutes, from the class kit.
• Teach the live class “Trace how preference pairs train a reward model”, 90 minutes, from the class kit.
• Show live a comparison where both responses have flaws: weigh a factual error against a missed instruction, decide how strong the preference is and record why.
• Walk learners through what happens to a preference pair: a reward model learns from many comparisons and guides reinforcement learning, while methods such as DPO train on the pairs directly.
• Help learners who flip their choice between similar pairs, call a tie to avoid deciding, or favour the longer response by default.

What to expect

• The classes are live, online and in English, and learners join from many countries. In the application you can list other languages you teach in.
• For each class you get a kit with a minute-by-minute plan, slides and speaker notes, plus a demo of a pairwise comparison and a recording of it as backup.
• You work remotely on your own computer and need a stable internet connection, a camera and a microphone.
• Each class has an instructor test, and you pass it before your first class.
• Before a teaching assignment starts, you sign an instructor agreement. It covers pay, confidentiality and how class material may be used.
• Each invitation states the class times and the price learners pay, and you see both before you accept.
• Already have a course on preference ranking and RLHF, or want to make one? You can sell your own recorded course on YPAI Learn too. Bring a course you've built, or build one with our kit. It's the same application.

What you bring

• Preference comparisons you wrote on public or made-up prompts, with your reasons, that you can walk through.
• Experience teaching, leading calibration sessions for raters, mentoring reviewers or explaining machine learning to non-specialists.
• Patience with learners who find it hard to commit to one response when both look acceptable.

You need

• Hands-on pairwise comparison or preference ranking work on real AI training or evaluation projects.
• A working grasp of how RLHF and related preference methods use human comparisons, enough to explain it in plain words.
• Fluent spoken English, the language the classes are taught in.
• 18 years or older.

Good to know

• Does your employer need to approve outside work under your contract? Then get that approval before you accept an assignment.
• The amount of teaching follows demand for these classes, and each invitation tells you the expected time commitment.
• There is no fee to apply, and no YPAI Learn course to buy or take first.