AI Red-Teamer (English & Chinese)
Job Description
As AI rapidly integrates into global societies, ensuring its safety and ethical alignment across diverse cultures is non-negotiable. Your critical eye and deep understanding of Chinese and English linguistic intricacies will be instrumental in safeguarding AI systems from unintended biases and harmful behaviors, directly impacting user trust and model integrity. This is a vital role in advancing responsible AI development.
Key Responsibilities
Develop and execute sophisticated adversarial attacks and test cases to expose safety gaps in AI models, specifically targeting Chinese-English bilingual interactions.
Identify and categorize instances of cultural insensitivity, political bias, misinformation, or harmful content generated by AI in both languages.
Provide comprehensive, actionable feedback and detailed reports on identified vulnerabilities, including linguistic and cultural context.
Contribute to the development of robust red-teaming frameworks tailored for East Asian linguistic and cultural specificities.
Analyze model responses for nuanced semantic errors, pragmatic failures, and potential for misuse across Chinese dialects and English.
Ideal Qualifications
Native or near-native fluency in Mandarin Chinese (including simplified/traditional characters) and English, with profound cultural awareness.
Proven experience in quality assurance, linguistic validation, or security testing, ideally with AI/ML systems.
Strong analytical skills to dissect complex AI outputs and identify subtle biases or safety risks.
Familiarity with common adversarial attack techniques against LLMs (e.g., prompt injection, data poisoning).
Ability to work independently and meticulously document findings in a structured manner.
Background in computational linguistics, sinology, or information security is a significant asset.
Project Timeline
Start Date: As soon as possible
Duration: 6 months, with potential for extension
Commitment: Part-time, 20-30 hours per week
Help us build safe, ethical, and culturally intelligent AI for a global audience!