A research team will compare outputs from several generative models using a fixed prompt and wants a group of scientists to assess the outputs. Which AWS solution supports human evaluation of model outputs?

Amazon Bedrock offers model evaluation capabilities, including human-in-the-loop workflows, which are ideal for having a group of scientists assess generative model outputs. This allows for subjective and nuanced feedback that automated metrics might miss. Automatic evaluation on Amazon Personalize is incorrect because Personalize focuses on recommendation systems, not general generative model evaluation. Content moderation on Amazon Rekognition is incorrect as Rekognition is for image and video analysis, primarily for content safety, not for evaluating generative text or other model outputs. Sentiment analysis on Amazon Comprehend is incorrect because Comprehend analyzes text for sentiment and entities, which is a specific NLP task, not a platform for broad human evaluation of generative models.