Earth Science Benchmark Task Designer

As an Earth Science Benchmark Task Designer, you will support AI training work based on Earth science, climate, and geospatial research datasets. You will use your subject knowledge to review model responses and create clear, evidence-based feedback.

The work is remote and contractor-based. Prior AI training experience is not required, but you should understand Earth science benchmark task design and be comfortable making consistent judgments from detailed project instructions.

• Evaluate model outputs using project rubrics.
• Review source materials and model outputs against project guidelines.
• Explain your decisions clearly so feedback can improve model behavior.
• Keep quality and judgment consistent across repeated review tasks.

What it pays and takes

This role is open worldwide. The listing includes a per-label payment and a time commitment, along with a required period of overlap with Pacific Time.

• Pay: $300 per completed label.
• Time commitment: Project details list 20+ hours per week; the role description asks for at least 40 hours per week.
• Schedule: Four hours of overlap with PST are required.
• Language: Fluent English.
• Experience: Entry-level; prior AI training experience is not required.
• Skills: Earth science benchmark task design knowledge, strong written communication, careful attention to detail, and the ability to follow detailed instructions and apply rubrics consistently.
• Work style: Able to work independently on remote contractor tasks.

How it works

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

About AI training work

AI training is the human work behind systems that learn from examples and feedback. People review model responses, prepare datasets, and explain what a system got right or wrong, so strong subject-matter judgment is valuable even when the work is entry-level in AI training.