About the Project

Pareto.AI is a human data-collection platform connecting leading AI researchers with trusted industry experts to collaborate on AI alignment, safety, and training projects. We are partnering with a frontier AI lab to evaluate an AI model's ability to complete realistic, complex spreadsheet work.

What You'll Do

• Design complex, realistic spreadsheet tasks that surface common failure points in AI-generated Excel work
• Test whether the model can complete tasks correctly, and iterate until the desired difficulty level is reached
• Identify and stress-test failure modes such as destructive formula changes, broken cross-references, incorrect structured table conversions, and loss of conditional formatting
• Write programmatically verifiable prompts with clear grading instructions for output files
• Top performers advance to QA roles ($100/hr), reviewing and validating tasks designed by others

Who We're Looking For

• Advanced proficiency in Excel — complex formulas, structured tables, conditional formatting, cross-sheet references (required)
• Ability to write clear, precise instructions that can be evaluated programmatically
• Python experience is a plus, but not required
• Professional background in Project Management, Operations, Consulting, Marketing, Data Analysis, or Accounting is preferred

Ideal Background

• Power Excel users who work with sophisticated spreadsheets day-to-day
• Analysts, consultants, or operations professionals who have encountered (and fixed) broken or fragile spreadsheet logic
• Anyone with a sharp eye for edge cases and a knack for breaking things intentionally

Why Join

• Contribute to cutting-edge AI safety and capability evaluation research
• Fully asynchronous, remote work — set your own schedule with 10+ hours/week commitment
• Clear path to advancement: strong task designers move into higher-paid QA roles
• Long-term project opportunity pending successful completion of the pilot phase
• Compensation: $75/hr for task design | $100/hr for QA

Project Timeline

Pilot runs through mid-March 2026, with the main phase to follow. Onboarding is rolling — apply now to be considered for the current cohort.

Apply

To apply, submit your CV. We review every application personally — no automated screening. If your background is a strong fit, you'll receive a direct link to join the project and complete your application.