What is AI training work, and who does it?

The short answer

AI training work is paid work that teaches AI models to be right. Labs hire people who already know a subject, a doctor, a lawyer, a physicist, a native speaker, to write the material models learn from and to judge what the models produce. You are paid for your judgement, usually by the hour, and you do the work remotely, in your own time. There are 853 such roles open on Tier1 right now, with a median rate of $70 an hour.

What the work actually is

Almost every project is a mix of the same few tasks:

• Evaluating answers. You read what a model wrote and grade it: is it correct, complete, safe, well reasoned. Often you compare two answers and say which is better and why. This is the work people mean by RLHF, reinforcement learning from human feedback.
• Writing reference answers. You write the answer an expert would give, so the model has something to learn from and be measured against.
• Writing hard problems. You set questions in your field that current models get wrong, with a verified answer. These become the benchmarks labs use to measure progress.
• Writing rubrics and guidelines. You define what a good answer looks like, so other raters can grade consistently.
• Recording and reviewing language. For speech and language projects: recording your voice, checking transcripts and translations, rating how natural a model sounds.

Who gets hired

The roles that pay well hire for a profession, not for AI experience. Nearly every role asks about your field and nothing about machine learning, because what the lab needs is the judgement you already have.

By profession, there are pages for accountants (10), voice and audio (45), physicists (27), chemists (12), biologists (13), mathematicians (8), phd holders (36), writers and editors (11), data scientists (18) and cybersecurity (14).

How it is organised

Labs rarely hire trainers directly. They work through platforms that recruit experts, run the projects and pay the people on them. You apply to a role on the platform's site, usually pass a screening step (a short interview or a test task in your field), and once accepted you work on the platform's own tools, on your own schedule, within the hours the project offers.

Projects start and end as labs need them, so the work comes in waves. Many people keep a profile on more than one platform so that there is always something open.

What it pays

Right now the median hourly rate across open roles is $70, and 175 roles pay $100 an hour or more. The pay guide breaks that down by field, language and pay type.

Is it for you?

It suits you if:

• you have real expertise in a field, or speak a language natively;
• you can explain your reasoning clearly in writing;
• you are careful: a missed error is the one thing the work cannot afford;
• you want flexible, part-time, remote work rather than a salary.

Before you apply anywhere, read how to tell a real project from a scam, then browse the roles by field.