Machine Learning Research Evaluation Expert

Create realistic machine learning research problems based on your research expertise, then solve them and write clear reference answers. Your work helps evaluate AI models by showing the difference between sound research judgment and answers that seem plausible but are wrong.

• Design problems about experiment design, ablations, model evaluation, and failure analysis.
• Explain the reasoning behind each step of your reference answers.
• Spot methodological mistakes and weak research conclusions that may take practical experience to recognize.
• Focus on evaluation work in your research specialty rather than completing a high volume of tasks.

What it pays and takes

• Pay: $115 per hour, paid weekly via Stripe.
• Schedule: At least 20 hours per week. The role is asynchronous with no fixed hours.
• Work arrangement: Remote contractor role, open to candidates based in eligible countries.
• Language: Clear written English and the ability to explain technical results step by step.
• Education: A master's or PhD in computer science, statistics, mathematics, or a related field.
• Experience: At least three years of full-time machine learning research after completing your degree. Postdoctoral research counts; doctoral study years do not.
• Skills: Hands-on experience designing experiments, running ablations, or evaluating machine learning models; strong Python skills; and experience with a deep learning framework such as PyTorch or JAX.
• Helpful experience: Publications at venues such as NeurIPS, ICML, ICLR, ACL, or CVPR, or work in an industry research lab or university research group.

How it works

Start your application on OpenTrain with your resume, then complete the application on the hiring site.

About AI training work

AI training work is the human work behind building and evaluating AI systems, including writing and reviewing examples that help measure model performance. Research expertise matters for this project because the evaluation problems need to distinguish sound methods from convincing mistakes.