Choose the kind of work first
Platforms list the same task under different titles, so it is faster to start from what you would do all day. Each hub shows the listings open now and the typical range of stated hourly rates.
-
RLHF evaluation and AI rating
Compare model answers, score them against a rubric, explain the call.
479 open · $30–$100/hr
-
Agentic tasks and RL environments
Build multi-step tasks and graders that agents are trained on.
69 open · $65–$85/hr
-
Red-teaming and AI safety
Try to make a model break its rules, then write up how.
112 open · $26–$60/hr
-
Transcription and speech data
Transcribe, segment and label audio, often in your native language.
75 open · $23–$23/hr
- Match your background to open roles
Pick your field and country, no account needed.
- Hiring now
Roles platforms flag as urgent, newest first.
- Platform reviews
Pay, payout schedule and screening for each platform we list.
- Compare two platforms
Side by side on pay, countries and how you get in.
- How to find AI training jobs
Where listings come from and which ones never reach job boards.
Prepare for the screening before you apply
Most platforms read your resume first, then run a skills test or an AI interview.
- Interview question bank
Screening interview questions with written answers, by professional background.
- Mercor AI interview
Format, timing and what the AI interviewer follows up on.
- micro1 interview with Zara
What the voice interview covers and how long it runs.
- Alignerr AI interview
The screening steps before your first project.
- Ethos expert call
The short AI call that places you on expert projects.
- Resume skills for annotation work
The skills and wording screeners look for.
Rezi.AI
Match your resume to each listing
Rezi rewrites your resume bullets to use the keywords in the job listing.
AI Trainer Academy
Learn the skills screening tests check
Free, self-paced modules on RLHF, data annotation, prompt writing and AI alignment.
Interview Coder
Get code suggestions during interviews
A desktop app that suggests code during technical interviews, in Python, Java and C++. Most AI training platforms ban outside help on their assessments, so check the rules first.
Learn how labs judge model answers
Rating tests check whether you apply a written standard the way the lab does. The labs publish most of those standards, and reading them is the best preparation.
- What AI training work is
The main task types and who does them.
- Seven task examples, walked through
What a rating, ranking or rewrite task looks like on screen.
- How rubrics work
How graders turn a rubric into a score, with examples.
- Red-teaming explained
What adversarial testing asks of you and where it pays.
- Glossary
RLHF, SFT, golden sets and the rest of the vocabulary in listings.
What the labs ask raters for
- Google Search Quality Rater Guidelines PDF · Google
The full manual Google gives its search raters, including how to judge page quality and trust.
- OpenAI Model Spec OpenAI
How OpenAI says its models should behave, rule by rule, with examples of good and bad answers.
- Claude's constitution Anthropic
The principles Anthropic trains Claude on, useful context for safety and preference tasks.
Where your ratings go
- Illustrating RLHF Hugging Face
The clearest plain-language walkthrough of where human ratings go.
- Training language models to follow instructions (InstructGPT) arXiv · 2022
The 2022 paper behind training chat models on human rankings of their answers.
- Training a helpful and harmless assistant with RLHF arXiv · 2022
Anthropic's paper on the comparison data raters produce.
- Constitutional AI arXiv · 2022
How Anthropic replaced some human harm labels with AI feedback guided by written principles.
Practice on real data and tools
- Arena (formerly LMArena) Free
Vote between two anonymous model answers. It is the same judgment a ranking task asks for.
- Anthropic HH-RLHF dataset Free · Hugging Face
Real chosen and rejected answer pairs. Read a few dozen before a rating test.
- Label Studio Free, open source
Open-source labeling tool for practicing text, image and audio annotation on your own data.
- CVAT Free tier
Open-source image and video annotation with boxes, polygons and object tracking.
- OWASP Top 10 for LLM applications OWASP
The standard list of LLM attack types, a common reference on red-teaming projects.
- MITRE ATLAS MITRE
A catalog of real attacks on AI systems, organized by tactic.
Plan for weeks when queues go quiet
Projects end without notice and a new one can take weeks to open. The usual fix is two or three active platforms plus a second source of freelance work.
- Working across several platforms
How full-time trainers stack platforms so one empty queue does not stop income.
- Earnings calculator
Weekly and monthly pay for the roles open to you, in your currency.
SolidGigs
Find freelance work for quiet weeks
SolidGigs emails you a short list of freelance jobs each day, picked from 100+ job sites.
Set aside tax as a contractor
Platforms pay you as an independent contractor and withhold nothing, so the tax is yours to estimate and file. Hardware, internet and a home office can count as deductions.
- Taxes on AI training income
Self-employment tax, quarterly payments and the 1099-NEC threshold, for US contractors.
- 1099 and contract roles
Open roles paid as an independent contractor.
- IRS Self-Employed Individuals Tax Center US
Quarterly estimated tax, self-employment tax and deductions.
- IRS Form 1099-NEC US
The form US platforms send when they pay you as a contractor.
- IRS Form W-8BEN Outside the US
US platforms ask non-US contractors for this form to confirm you are not a US taxpayer.
- GOV.UK: working for yourself UK
When to register as self-employed and file a Self Assessment.
Keeper Tax
Find tax write-offs in your spending
Keeper scans your bank transactions for work expenses you can deduct.
Check a platform before you apply
Real platforms verify your ID and never charge you to start. Fake recruiters copy their names and job titles.
- Is AI training legit?
Why real platforms ask for ID, and the signs of a fake recruiter.
- Interview copilots and getting caught
How platforms detect assistance during assessments.
- FTC: job scams US government
Fake check and pay-to-start schemes, and where to report one.
See the data behind our numbers
- Research
Pay surveys and listing analyses, with the data published.
- AI Training Pay Index
Monthly pay and hiring figures, each with its own link to cite.
- How we count pay
Which rates count as stated, and which are estimates.
- Open data
Download the datasets behind our pages.
Some tools on this page are partner links. How partner links work