Remote AI training and data labeling jobs

Computational biologists with a PhD or equivalent design agentic life-science tasks where AI must navigate data, run analyses and troubleshoot to a valid result. $300 per approved task, 40+ hours a week, 5-week contract, 4-hour Pacific Time overlap.

Physicists, chemists and materials scientists turn real workflows (simulation, spectroscopy, data fitting, molecular analysis) into reproducible terminal benchmark tasks for AI agents. $300 per approved task, 40+ hours a week, 5-week contract.

Mathematicians and statisticians with a PhD or equivalent build terminal tasks (optimisation, ODEs, stochastic modelling, numerical methods) with rigorous graders for Terminal Bench Science. $300 per approved task, 40+ hours a week, 5-week contract.

Earth scientists with a PhD or equivalent build terminal-based benchmark tasks (geospatial, climate, ocean, geology) that test AI agents on Terminal Bench Science. $300 per approved task, 40+ hours a week, 5-week contract, 4-hour overlap with Pacific Time.

Computational life scientists design multi-step tasks that AI agents must solve by writing and running code on scientific data. Paid $280 per approved task, 5-week contract, 40+ hours a week with 4 hours of PST overlap.

PhD-level biologists design graduate-level problems and reference answers for LifeSciBench, a benchmark for AI scientific reasoning. Non-coding. Full-time hours (40/week) for one month with a PST overlap.

PhD-level bioinformaticians design hard agentic genomics tasks, datasets, reference solutions and graders for GeneBench Pro. Contract, 5 weeks, at least 40 hours a week with PST overlap. $260 per approved task.

Biology graduates author linked scientific coding problems with tested Python solutions for the SciCode training set, which the ad says is run in partnership with NVIDIA. 8-week contract, 40 hours a week, 12 countries. No rate published.

Chemistry graduates write multi-step scientific coding problems and verified Python solutions with unit tests for Turing's SciCode AI training set. 8-week contractor assignment, 40 hours a week, 12 countries. No rate published.

Mathematics graduates write chained scientific coding problems with verified Python solutions and unit tests for Turing's SciCode set. 8-week contractor assignment, 40 hours a week, 12 countries. No rate published.

Physics graduates author multi-part scientific coding problems with verified Python solutions and unit tests for Turing's SciCode AI training set. 8-week contractor assignment, 40 hours a week, open only to a list of 12 countries. No rate published.

Graduate-level STEM experts write and solve hard problems that expose weaknesses in large language models, give step-by-step reasoning, and help define evaluation benchmarks. Remote contractor assignment, open to current Master's, PhD and postdoc candidates. No rate published.

Biology graduates write and solve hard problems with step-by-step reasoning to probe large language models and shape new benchmarks. Remote contractor assignment; no rate, hours or duration published.

Chemistry postgraduates create and solve advanced problems with text, equations and diagrams to fine-tune language models. Remote contractor assignment; no rate, hours or duration published.

Physics graduates design and solve hard physics problems to probe language models, write step-by-step solutions and help define benchmarks from undergraduate to PhD level. Remote contractor assignment, open to current Master's and PhD students. No rate published.

Mathematicians write original hard problems, solve and review model solutions, and verify proofs in Lean to evaluate LLMs. Master's or PhD students and graduates welcome. 20 to 40 hrs/week, 4 hrs overlap with PST.

Analytical thinkers write math and logic problems that language models get wrong, then explain the correct answers step by step to help fine-tune the models. Full-time 40 hours a week, one-month contract with PST overlap and no specialist domain experience needed. No rate published.