Bilingual Turkish Generalist Expert — AI Safety

What you would do

• Compose challenging Turkish prompts targeting sensitive subject areas to test model behavior and boundary responses
• Apply structured evaluation frameworks to classify prompts and assess the quality of model outputs
• Identify manipulation attempts, escalation tactics, and problematic language patterns in Turkish interactions
• Document evaluation decisions with clear reasoning supporting each classification and flag determination
• Provide detailed feedback to research teams on model weaknesses and safety gaps

Who they want

• Native or near-native Turkish fluency with professional written English capability
• Bachelor's degree (completed or actively pursuing)
• Demonstrated ability to reason clearly in writing with meticulous attention to procedural detail
• Sound judgment in evaluating sensitive topics and potential dual-use information
• Optional: Background in content review, adversarial testing, or trust and safety work

Main skills

What the interview asks about

1. 1.Turkish adversarial pattern recognition

   Spotting manipulation tactics unique to Turkish requires linguistic expertise and cultural understanding essential to accurate safety classification.

   For example: “A Turkish user employs increasingly aggressive persuasion strategies to bypass safety guardrails in Turkish. Describe how you would identify escalation patterns and document your classification reasoning.”

2. 2.Bilingual prompt design

   Creating challenging yet ethical Turkish prompts demands balancing authentic language with responsible evaluation frameworks, directly supporting model safety work.

   For example: “Construct three progressively challenging Turkish prompts on a politically contentious issue that test model consistency without becoming harmful. Explain your reasoning for each variation.”

3. 3.Dual-use information judgment

   Navigating decisions about sensitive content requires contextual judgment that respects safety objectives while understanding Turkish and regional perspectives.

   For example: “You encounter a Turkish prompt about a potentially legitimate technology with dual-use applications. How would you classify it and what additional context would influence your assessment?”

4. 4.Framework application consistency

   Applying structured classification guidelines reliably ensures quality control across large evaluation volumes and maintains standardized safety criteria.

   For example: “Given five Turkish interactions ranging from low to high sensitivity, apply the classification framework to each case and explain any borderline categorizations.”

5. 5.Written reasoning clarity

   Clear documentation of assessment rationales creates transparency in safety evaluations and enables continuous refinement of evaluation criteria.

   For example: “For a flagged Turkish prompt, write a complete reasoning document covering the risk classification, specific problematic elements, and escalation justification.”

A task you may get

Candidates receive a set of Turkish prompts spanning risk levels with the evaluation framework and must categorize each, identify concerning patterns, and provide detailed written reasoning for decisions.

How to prepare

• Research core AI safety concepts, content moderation terminology, and dual-use information definitions to build foundational knowledge
• Study Turkish cultural and social contexts to inform judgment when evaluating sensitive content related to regional topics
• Review sample classification annotations and reasoning documents to understand framework application standards
• Gather examples of sensitive Turkish-language discussion topics and practice constructing prompt variations

The facts

Pay

$23–27/hr

Hours

Part time, 7 hours a week

Where

Remote · Remote — MENA preferred

Field

Miscellaneous

Posted

9/4/2026

Places left

1

We wrote this page from the public Mercor listing. It may be out of date, so read the full posting before you apply.