$48–52/hr · Mercor · Part time, 7 hours a week
A bilingual PhD scientist develops Croatian-language prompts and evaluates model responses to enhance AI safety in technical domains.
What you would do
- Author technically sophisticated prompts in Croatian designed to probe model capability and safety on specialized topics
- Evaluate model-generated answers for scientific correctness, quality, and appropriate treatment of dual-use subject matter
- Classify technical prompts and conversations according to established safety and quality frameworks
- Identify responses that mishandle sensitive information or contain scientific inaccuracies
Who they want
- PhD in chemistry, biology, or closely related discipline possessing sufficient depth to create expert-level prompts in your area
- Native or near-native Croatian fluency combined with business-standard English written competence
- Extensive knowledge of contemporary laboratory techniques and computational approaches within your specialty
- Well-developed judgment regarding scientific safety risks and responsible stewardship of sensitive technical knowledge
Main skills
What the interview asks about
1.Domain technical depth
You must identify when models provide scientifically inaccurate responses; insufficient expertise allows errors to slip through uncaught.
For example: “Consider an organic chemistry question about synthesis methodology. What three realistic errors could a model make, and how would your expertise distinguish them from correct approaches?”
2.Croatian scientific communication
Your prompts form the training dataset in Croatian; imprecise or poorly-structured prompts create low-quality training and confuse model learning.
For example: “Compose a Croatian-language prompt about molecular biology that tests whether a model grasps genetic engineering methodology. Which technical terms and concepts did you emphasize and why?”
3.Dual-use risk assessment
Recognizing when information crosses from legitimate science into potentially dangerous misapplication defines your core safety contribution; missing this defeats your purpose.
For example: “A model provides detailed procedures for fermentation techniques. How would you determine if the response appropriately balances scientific education against enabling risky applications?”
4.Prompt quality and testing value
Poorly designed prompts fail to challenge models; skillfully constructed ones reveal gaps in model understanding.
For example: “Design two distinct Croatian prompts in microbiology that test different facets of model capability. Explain what makes each prompt effective at revealing model weaknesses.”
5.Cross-disciplinary risk recognition
Danger often emerges when knowledge from one domain combines with techniques from another; identifying these patterns separates expert contributors.
For example: “You specialize in radiochemistry. How would safety considerations shift if models were generating responses combining your specialty with engineering or pharmaceutical chemistry knowledge?”
A task you may get
Write 2-3 Croatian-language scientific prompts in your specialization and explain what comprehensive responses should address, plus identify likely areas where model reasoning might fail.
How to prepare
- Review cutting-edge scientific literature in your discipline, noting common misconceptions where models might produce incomplete or incorrect answers
- Prepare examples from your field where technical accuracy requires distinguishing nuance from oversimplification, and potential harmful simplifications to watch for
- Practice bilingual technical communication, particularly translating specialized concepts between English and Croatian
- Consider safety-sensitive topics in your field and how responsible guidance differs from dangerous information
The facts
- Pay
- $48–52/hr
- Hours
- Part time, 7 hours a week
- Where
- Remote · Remote — Eastern Europe 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.