March 2023 Summaries

Multimodal moderation is an advanced approach to content moderation that addresses the limitations of traditional text-based methods by integrating various modalities such as text, images, video, and audio. This approach leverages artificial intelligence techniques like natural language processing, computer vision, and speech-to-text, alongside human expertise, to provide a more nuanced and accurate analysis of content. It is particularly effective in uncovering harmful messages that are conveyed through multimedia formats and might be missed by text-only moderation. The benefits of multimodal moderation include increased detection accuracy, reduced bias, improved efficiency, and adaptability to new content forms and behaviors. Recent research from the University of Toronto highlights its effectiveness, particularly in identifying implicit hate speech in multimedia content. Implementing multimodal moderation involves identifying relevant modalities, collecting and labeling training data, training AI models, and integrating them into platforms' moderation systems with ongoing refinement. This approach ensures a safer online environment by providing a comprehensive understanding of user-generated content, thus enhancing the safety and well-being of platform users.

Clarifai has introduced several updates aimed at enhancing user experience and functionality on its platform. These updates include the addition of a single sign-on feature allowing users to register using their Google accounts, a modernized form for creating deep-trained models, and improvements in user interface design, such as updated titles for resources and a redesigned pending verification page. The platform now supports infinite scrolling and pagination for concepts and collaborations, respectively, and offers improved search functionalities and video interface controls. Additionally, users can switch between multiple GPU devices for faster model training and have the ability to overwrite model output information in workflow graphs for more tailored use cases. Several bug fixes have been implemented, such as addressing issues with search execution timing, model viewer loading, and deep train configuration resets. The platform encourages engagement through its Slack group for users to share insights and connect with engineers.

Artificial Intelligence (AI) is increasingly utilized across various industries, and creating AI-driven applications often requires the incorporation of images. For developers needing imagery, there are several accessible options. Free stock sites such as Unsplash, Pixabay, and Pexels offer thousands of royalty-free images suitable for commercial use, which can be searched and filtered by keywords or categories. Social media platforms like Facebook, Instagram, Twitter, and Pinterest also provide a wealth of user-posted images, though developers must ensure they are free from copyright restrictions before use. Google Advanced Image Search offers another avenue, allowing users to refine image searches by type, size, color, and usage rights. For those seeking complete ownership and privacy, using personally taken photographs is ideal, as these images are free from copyright concerns and can be selectively exported based on content using tools like Google Photos. By leveraging these resources, developers can efficiently source the right images for their AI applications.