Safe-CLIP - Removing NSFW Concepts from Vision-and-Language Models

Learn how to remove NSFW (Not Safe For Work) concepts from vision-and-language models through this 23-minute research presentation from the University of Central Florida. Explore the Safe-CLIP methodology, which addresses the critical challenge of filtering inappropriate content from multimodal AI systems that process both visual and textual information. Discover the technical approaches used to identify and eliminate problematic concepts while maintaining the model's performance on legitimate tasks. Examine the evaluation metrics and datasets used to measure the effectiveness of content filtering, and understand the broader implications for deploying safer AI systems in production environments. Gain insights into the intersection of AI safety, computer vision, and natural language processing as researchers tackle the important problem of content moderation in modern machine learning models.