$24–28/hr · Mercor · Part time, 7 hours a week
Evaluate AI responses to advanced scientific questions in Thai for accuracy, safety, and responsible handling of dual-use research topics.
What you would do
- Compose sophisticated questions in Thai on chemistry, biology, or related domains to test model reasoning
- Evaluate AI responses for scientific accuracy, methodological soundness, and appropriate scope
- Annotate responses for helpfulness, sensitivity to dual-use concerns, and responsible disclosure practices
- Apply structured classification guidelines to categorize technical prompts and conversations
- Provide feedback distinguishing legitimate research questions from concerning requests
Who they want
- PhD (completed or ongoing) in chemistry, biology, or closely related field with depth across specialized subdomains
- Native or near-native Thai fluency plus business-level written English for technical communication
- Hands-on experience with current lab methods and software tools specific to your specialty
- Sound judgment about scientific safety and responsible handling of dual-use research information
- Minimum 20 hours weekly commitment; based in Thailand or Southeast Asia preferred but not required
What the interview asks about
1.Writing expert-level prompts in Thai
Training data quality depends on scientifically sophisticated prompts that reflect real researcher thinking, so shallow or oversimplified questions produce models that fail specialists.
For example: “Write a prompt in Thai asking an AI to explain the mechanism by which a specific synthesis pathway minimizes a particular safety hazard. Demonstrate both technical accuracy and clarity.”
2.Identifying dual-use information risks
AI safety requires catching when models provide information that, while technically accurate, could enable harmful applications that deserve restricted disclosure.
For example: “An AI provides step-by-step details on a technique that would typically require safety equipment and institutional oversight. How would you evaluate whether the model's response level is appropriate?”
3.Evaluating scientific methodology
Models trained on flawed or outdated methodology produce dangerous recommendations, so your ability to spot procedural errors is critical for safety outcomes.
For example: “An AI explains a lab procedure but omits a critical safety step or suggests an outdated technique. Walk through how you'd identify the gap and explain why the omission matters.”
4.Communicating across languages
Training value depends on your ability to articulate domain expertise clearly in both Thai and English, ensuring classification decisions are interpretable to research teams.
For example: “Classify a Thai-language prompt as concerning based on dual-use considerations. Explain your reasoning in English in a way that helps non-specialist researchers understand your safety concerns.”
A task you may get
Write two expert-level prompts in Thai on chemistry or biology topics, evaluate AI responses for scientific accuracy and dual-use safety, and provide written feedback on findings.
How to prepare
- Refresh your understanding of current safety frameworks in your field including regulatory requirements and institutional oversight
- Review dual-use research governance and common examples of restricted-information handling in academic contexts
- Study technical terminology in both Thai and English to ensure you can classify and communicate precisely
- Familiarize yourself with red-teaming approaches and how to recognize potentially problematic model outputs
The facts
- Pay
- $24–28/hr
- Hours
- Part time, 7 hours a week
- Where
- Remote · Remote — Southeast Asia preferred
- Posted
- 9/4/2026
We wrote this page from the public Mercor listing. It may be out of date, so read the full posting before you apply.