Remote AI Evaluation Researcher, Part-Time

OpenTrain AI, Inc.; United States; Remote; USD 41,000 - 83,000; Part time

Job summary

OpenTrain AI, Inc. seeks contractors to create difficult, evidence-based Q&A materials for advanced AI model evaluation. You will earn $30 to $60 per hour equivalent, with fewer than 20 hours per week, on a remote, part-time basis.

Applicants must have a PhD or active candidacy and deep subject-matter expertise, with fluent written English. The role involves researching primary literature, documenting citations, and writing high-quality questions and defensible answers.

Qualifications

• Fluent written English required.
• Completed PhD, active candidacy, or equivalent research experience.
• Deep subject-matter expertise and familiarity with primary literature.
• Experience finding and comparing information from authoritative references.
• Ability to write challenging, original, methodologically sound questions.

Responsibilities

• Create high-difficulty questions and defensible answers in your area of expertise.
• Research primary literature and authoritative references, documenting citations and reasoning.
• Design questions requiring advanced reasoning or synthesis across sources.
• Test questions against AI systems and refine items to resist shortcuts.
• Write clear, precise, and unambiguous question-and-answer pairs.
• Follow reviewer feedback and project guidelines and quality standards.

Skills

Analytical thinking

Written communication

Research judgement

Independent work

Education

PhD or doctoral candidacy

Job description

OpenTrain AI, Inc. seeks contractors to create difficult, evidence-based Q&A materials for advanced AI model evaluation. You will earn $30 to $60 per hour equivalent, with fewer than 20 hours per week, on a remote, part-time basis.

Applicants must have a PhD or active candidacy and deep subject-matter expertise, with fluent written English. The role involves researching primary literature, documenting citations, and writing high-quality questions and defensible answers.