Lead a pod of about 5–10 trainers building terminal-based materials science tasks for a frontier AI training and evaluation project: review tasks end to end (DFT, molecular dynamics, thermodynamics, crystallography), check scientific correctness and reproducibility, review automated graders and mentor trainers.
Lead a pod of trainers building terminal-based chemistry tasks for Terminal Bench Science: review tasks end to end for chemical correctness, reproducibility, safety and grader reliability (units, convergence, stereochemistry, force fields), and coach trainers.
Lead a pod of trainers building terminal-based physics tasks for Terminal Bench Science: review every task end to end (datasets, simulation inputs, environments, reference solutions, automated tests) for physical correctness, reproducibility and grader reliability, and coach trainers.
Lead a pod of 5–10 trainers building terminal-based mathematical tasks for Terminal Bench Science: review problem statements, environments, reference solutions and automated tests for correctness, stability and reproducibility, design graders and mentor trainers.
Non-coding remote project: evaluate AI responses in single- and multi-turn conversations against safety policies, give 1–5 ratings and write short evidence-based justifications. Content includes sensitive topics such as self-harm, violence and abuse. Hourly pay; 1–2 weeks with possible extension, starting around 23 October.
Remote AI audio project for music professionals: listen to short clips and identify the instrument family and the specific instrument, following annotation guidelines and audio references. Getting started requires a short (about 15-minute) qualifying assessment; pay is in USD.
Five-week contract for computational life science experts: design realistic multi-step agentic workflows where AI agents analyse scientific data, write and run code and produce verifiable results, with reproducible expert solutions and grading criteria. At least 6 hours a day and 40 hours a week, with 4 hours’ overlap with U.S. Pacific time.
Remote quality-review role on an AI safety project for a US AI company: audit trainers’ evaluations for policy misapplication, incomplete reasoning, scoring drift, schema errors and inconsistent annotations, apply a detailed AI safety policy framework, give corrective feedback and document recurring error patterns. The listing states that the project includes sensitive content (self-harm, violence and abuse), with waivers and wellbeing support.
16-week remote AI evaluation project (with possible extension): review AI-generated prompts and responses in English that use selected data from your own Google apps, Google Messages and device SMS, and rate them for relevance, accuracy, helpfulness and personalization. This work uses your personal account data: read the privacy and consent terms on the official page before applying.
Remote project of 8–10 weeks for 3D and digital artists: create stylized 3D models and digital assets from briefs and references (low-poly, voxel, handcrafted and material-driven styles), recreate textures such as clay, fabric and paper, and produce consistent, optimized variations. The listing describes earnings as based on approved tasks, paid in USD, with work starting after portfolio approval and onboarding.