眠眠 姜

I review Chinese-language AI outputs for safety, accuracy, relevance, clarity, completeness, and instruction following. My work combines structured evaluation criteria with clear, concise feedback.

In educational content review, I checked materials, exercises, answer keys, and explanations before delivery to students. I identified incorrect answers, logical inconsistencies, unclear explanations, and formatting issues while applying predefined standards across high-volume content.

Through independent AI safety evaluation projects, I assessed prompts and responses for intent, adversarial framing, dual-use potential, escalation, severity, and risk category. I classified privacy, fraud, cybersecurity, harassment, and sensitive-information risks and recommended Allow, Partial, or Refuse outcomes with documented rationale.

I compare outputs from ChatGPT, Claude, Gemini, and DeepSeek, identify hallucinations and unsupported claims, and refine ambiguous prompts into practical review checkpoints. I also design quality-control and formatting workflows for Word-based educational documents, using AI to improve efficiency while manually verifying final results.