Multimodal Response Evaluator (Aether Project) – Outlier AI
In the Aether Project at Outlier AI, I evaluated AI-generated multimodal outputs to ensure cross-modal accuracy and user intent alignment. Assessments emphasized model safety, constraint adherence, and rapid fact-checking against references. My work contributed to maintaining high-quality outputs and detecting harmful content or hallucinations. • Evaluated text, image, and data visualization outputs for consistency. • Verified model adherence to safety and formatting constraints. • Flagged and documented potentially harmful or factually incorrect content. • Ensured user prompt intent was correctly identified by the model.