$19–23/hr · Mercor · Part time, 7 hours a week
Compose advanced Vietnamese science prompts and judge how well AI models respond in terms of accuracy and responsible handling.
What you would do
- Write sophisticated science prompts in Vietnamese across chemistry, biology, radiological, and nuclear specializations
- Assess model replies for technical accuracy, appropriateness, and sound reasoning
- Evaluate how models address sensitive or dual-use topics and whether they demonstrate safety awareness
- Categorize and organize prompts using provided structural frameworks and guidelines
- Document areas where model responses demonstrate strength and areas needing refinement
Who they want
- PhD in chemistry, biology, or adjacent fields (defended or currently in progress)
- Native or near-native Vietnamese fluency paired with fluent professional English
- Hands-on laboratory experience and familiarity with contemporary research methodologies
- Clear ethical judgment regarding sensitive science and potential risks from misuse
- Ideally: specialization in synthesis, microbiology, radiochemistry, energetic materials, or molecular genetics
Main skills
What the interview asks about
1.Crafting scientifically rigorous prompts
The quality of your prompts determines what you can learn about how models reason; weak prompts waste evaluation effort.
For example: “Imagine a biology prompt about genetic techniques. You want to test whether the model understands both CRISPR mechanisms and the ethical frameworks that govern their use. How would you frame this query in Vietnamese to elicit substantive, nuanced responses?”
2.Evaluating accuracy beyond surface correctness
A response can be factually right but miss important context or represent outdated understanding.
For example: “A model explains a biochemical pathway accurately based on textbooks from 15 years ago. Recent research has refined understanding of a key step. Do you rate the response as accurate, incomplete, or partially misleading? How do you document this?”
3.Navigating safety and sensitivity boundaries
Your judgment determines how strictly the system treats sensitive topics versus how openly it engages with legitimate research.
For example: “You have a model response about bacterial culturing that includes procedural details applicable to common lab work but also to sensitive applications. It doesn't explicitly hide anything, but context matters. How would you assess this response?”
4.Bilingual precision in complex domains
Scientific terms in Vietnamese can diverge from English counterparts in subtle ways; precision matters.
For example: “A model responds in Vietnamese using a term that in English would clearly mean X, but in Vietnamese scientific discourse has evolved to encompass X and Y. The model's usage is technically reasonable but ambiguous. Does this affect your rating?”
How to prepare
- Deepen your understanding of dual-use research frameworks in your scientific discipline and how to discuss them responsibly
- Develop 3-4 sophisticated Vietnamese prompts that would meaningfully test model reasoning in different specialty areas
- Consider how to evaluate scientific depth: a model could sound competent while oversimplifying or missing nuance
- Practice identifying when model responses reflect current scientific consensus versus outdated knowledge
The facts
- Pay
- $19–23/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.