Web Researcher – AI Benchmarking
Posted 9hrs ago
Employment Information
Report this job
Job expired or something wrong with this job?
Job Description
Web Researcher designing difficult, evidence-traced benchmarks for frontier AI browsing agents. Building reproducible open-web investigations and validation documentation.
Responsibilities:
- Design challenging open-web research problems for evaluating advanced AI browsing agents
- Start from objectively verifiable facts and construct questions that make those facts difficult to discover
- Build multi-step clue structures involving dates, people, places, organizations, works, events, records, and quantities
- Ensure every clue is independently verifiable and subject to clear constraints
- Research across government databases, institutional sources, archives, registries, and PDF documents
- Produce complete evidence trails citing exact pages, tables, sections, or records
- Record validation searches and document what obvious search approaches return
- Test whether research questions remain difficult across multiple search attempts
- Refine questions to eliminate ambiguity while preserving difficulty
- Deliver structured, reproducible research documentation
Requirements:
- Background in reference librarianship, archives, or special collections
- Experience in investigative journalism or professional fact-checking
- Experience with OSINT, due diligence, KYC, or investigative research
- Background in patent, prior-art, or legal-discovery research
- Experience with genealogy or historical records research
- Experience with competitive quizzing, puzzle design, or puzzle-hunt construction
- Familiarity with JSON or structured data formats
- Commitment of 40 hours per week, at least 4 hours per day, including 4 hours overlap with PST














