Freelance AI Trainer (OpenTrain / Outlier AI / Scale AI)
Text
RLHFProvided RLHF preference data by evaluating and ranking LLM outputs across dozens of projects to improve language model alignment. Conducted adversarial prompt engineering and red-teaming to surface jailbreak vectors, safety gaps, and hallucination behaviors in production systems. Performed reasoning step verification with process-level feedback to support downstream reward model training and quality improvement. • RLHF preference ranking from ranked model outputs • Red-teaming and safety testing for jailbreak and hallucination detection • Reasoning verification for math and logic chain steps • High-quality scoring and consistency across annotation platforms
2023 - Present