This Project Grant award from the National Science Foundation's (NSF) Computer and Information Science and Engineering (CISE) program (CFDA 47.070) provides $400,000.00 to Carnegie Mellon University to develop new methodologies for visualizing, understanding, and editing multimodal generative AI models. The project aims to increase the accountability and safe use of these AI models by addressing challenges such as unpredictability, bias, and the potential for generating harmful or copyrighted content. Key research thrusts include developing techniques to analyze the internal mechanisms and hierarchical structures of pre-trained multimodal generative models, exploring model editing algorithms to fix issues, and creating new attribution methods to assess the influence of training data on model outputs. The award period runs from September 15, 2024, to August 31, 2028, and the research results will be widely disseminated through tutorials, workshops, and courses to engage the research community and inform policymakers.