STEM Subject Expert Model Defeating Content For AI Training

Job title: STEM Subject Expert Model-Defeating Content for AI Training (Physics / Chemistry / Biology / Mathematics / Advanced STEM)

Location: Remote, India

Type: Contract, part-time or full-time. Minimum 15 hours per week.Experience range: Any

About the roleWe create model-defeating content for AI models built by leading US AI labs. Model-defeating content means original, hard questions that today’s best AI models answer incorrectly. Those are the questions that teach a model something new, so each question is valuable only if the AI fails it. Your job is to write such questions in your subject, with complete step-by-step solutions, and to grade the AI’s attempts. Think of it as setting an exam for a very strong student and finding the questions that catch them out. This is expert work, not data entry.
What you will do

• Write original, challenging questions in your subject at doctorate or postgraduate level, designed to expose gaps in AI reasoning: multi-step problems, subtle traps, edge cases and precise numerical answers.
• Test each question against the specified AI models. Only questions the models fail are accepted.
• Write full worked solutions with numbered steps and final answers in LaTeX.
• Grade AI-generated answers on accuracy, completeness and structure, and explain where and why they fail.
• Meet weekly quality and volume targets.
• Subjects Expertise: Any one or more of them - Physics, Chemistry, Biology, Mathematics, and other advanced STEM subjects such as Statistics, Computer Science, Electrical and Mechanical Engineering, and Biochemistry.

Who we are looking for

• Master’s or PhD in a STEM subject from IIT, IISc, NIT, AIIMS, a central university or another Tier 1 or Tier 2 institute. Final-year PhD scholars are welcome.
• Proven problem-setting ability: teaching, research, olympiad or competitive-exam coaching, or question-paper setting.
• Fluent in LaTeX and academic English.
• Strong reasoning and attention to detail. Correct units, signs and significant figures matter.
• Able to produce original work. All submissions are checked for plagiarism and AI use.

Selection process:

• Resume screen.
• One sample model-defeating question with solution in your subject, submitted in LaTeX.
• A 20-minute call.