Science & Research AI Training Jobs

Science AI training jobs pay researchers to write and verify the physics, chemistry, biology and math that models train on. Frontier labs are pushing models toward graduate-level scientific reasoning, and the ceiling on that is the supply of scientists willing to author problems, check derivations and grade proofs. That is what these listings buy.

Below is every live science and research listing from the six marketplaces in this feed, refreshed daily, with published rates.

150 open science & research listings live right now.

FiltersActive
Clear all filters
Minimum pay
$25+ / hour$50+ / hour$75+ / hour$100+ / hour

Filtering by rate hides per-task listings, which have no hourly equivalent.

Category
Language & audio197Software engineering150Science & research150Finance & accounting117Law105Business operations98Data & machine learning88Medicine & healthcare83Arts & design80Cybersecurity72Engineering (other)51Sales & marketing38
Experience
Entry level21Mid level160Senior198Lead / principal30
Commitment
Contract / hourly848Part time51Full time29Task based25
Arrangement
Remote936Hybrid14On site3
Marketplace
Alignerr486Mercor418micro1364DataAnnotation19Outlier15Terac11
Date posted
Past weekPast monthPast 3 months

What the work looks like

The core task: produce problems a model cannot pattern-match its way through, with worked solutions a grader can verify. That means exam-grade questions with rationales, derivations checked step by step, experimental-design critiques, and grading model attempts at published-paper reasoning. Math work runs from olympiad problems to real analysis; lab sciences lean on mechanism and methods questions.

The bar is graduate-level command of a field. PhDs and PhD students dominate the contributor pool, but a masters with research experience passes many screens. What the assessments filter for is rigor: whether your solution is airtight and whether you catch the subtle error planted in someone else’s.

Where these listings come from

Six marketplaces publish this work openly. Every listing above links to the one that posted it, and each has its own page here covering pay, screening and who gets hired.

MercorOutlierDataAnnotationAlignerrmicro1Terac

Science & research AI training jobs, answered

What do scientists do in AI training work?

Author original problems with worked solutions, verify other contributors’ solutions, grade model reasoning against rubrics, and stress-test models with questions where the plausible answer is wrong. Some projects are field-specific, organic synthesis routes, quantum mechanics derivations, genetics problem sets; others want general scientific-method judgment. It is the problem-setting side of academia, paid by the hour.

How much do science AI training jobs pay?

Published rates typically run $40 to $100 an hour, with math and physics problem authoring at the upper end and some specialist projects above it. That is competitive with adjunct and postdoc hourly equivalents by a wide margin, which is why the contributor pool is full of academics. Rates are on each card above.

Do I need a PhD?

For the best-paid authoring work, usually yes or nearly: labs want problems that challenge models trained on every textbook, and that takes research-level depth. Verification and grading roles accept masters-level scientists and strong undergraduates in math. Every screen is a work test, so demonstrated rigor can outrun the credential.

Can academics do this without conflicts?

Generally yes. The work is independent contracting on original problems, not university IP, and it does not involve your research data. Check two things: your institution’s outside-consulting policy, most allow a day a week, and grant terms if you are federally funded. The marketplace NDA covers the other direction: you cannot publish or reuse the problems you author.

How are scientists screened?

Credential verification plus a subject assessment: solve problems, find the planted error, write a solution to grading standard. Mercor adds a recorded AI interview where you reason aloud. Scientists fail these by writing for peers instead of for a rubric; the screens reward complete, legible reasoning at the level of a very good TA. That style is practisable before the real screen.

The screen is where these roles are won

Every marketplace above screens with an interview or assessment in your own domain before assigning paid work. Run that interview with Skillora first and get feedback on how you explained your reasoning, not just on what you said.

Practise the screen free

Browse other fields

Software engineeringData & machine learningMedicine & healthcareLawFinance & accountingLanguage & audioBusiness operationsAll AI training jobs