Web Research Benchmark Task Author

You will create realistic research tasks that test where advanced language models struggle. The work combines investigative research, source verification, benchmark writing, model testing, and quality review.

• Research niche topics and find factual answers that basic searches do not easily reveal.
• Cross-check evidence, resolve conflicting information, and document exact source URLs.
• Write clear benchmark prompts with precise answer keys supported by primary or highly credible sources.
• Run tasks against advanced language models to confirm the intended weaknesses.
• Inspect model reasoning, search paths, run logs, and intermediate outputs.
• Determine whether failures come from model limits, search behavior, source choices, or unclear prompts.
• Apply proven research patterns to new subject areas and review benchmark tasks from other trainers.

What it pays and takes

This is a part-time contractor role listed as entry level. The listing does not provide pay details.

• Time: 20+ hours per week.
• Location: Open worldwide.
• Language: Fluent in English.
• Experience: At least one year in investigative research, fact-checking, AI benchmark authoring, red-teaming, or technical quality review.
• Research skills: Intermediate to advanced web search and OSINT skills, including work with academic archives, local records, or obscure databases.
• Technical skills: Strong analysis of model reasoning, search logs, and source selection, plus clear technical writing for precise prompts and defensible answer keys.
• Education: A relevant bachelor's degree or higher in an engineering field.
• Benchmark knowledge: Familiarity with GAIA, BrowseComp, SimpleQA, SWE-bench, Terminal-Bench, or Humanity's Last Exam.
• Helpful experience includes prompt-injection or reasoning-trap task design, published investigative or academic research, work with difficult-to-find records, or leadership in AI evaluation quality review.

How it works

Apply on OpenTrain with your resume, then complete the application on the hiring site. OpenTrain is free to use and helps people find AI training work and build a profile of relevant experience.

About AI training work

AI training is the human work behind systems that learn from examples, including writing and rating model responses, testing model behavior, and evaluating search or research results. This role helps assess whether AI systems can find reliable evidence, follow research steps, and produce defensible answers.