Holistic Safety and Responsibility Evaluations of Advanced AI Models

arXiv:2404.14068v1 Announce Type: cross
Abstract: Safety and responsibility evaluations of advanced AI models are a critical but developing field of research and practice. In the development of Google DeepMind's advanced AI models, we innovated on and applied a broad set of approaches to safety evaluation. In this report, we summarise and share elements of our evolving approach as well as lessons learned for a broad audience. Key lessons learned include: First, theoretical underpinnings and frameworks are invaluable to organise the …

abstract advanced advanced ai ai models arxiv cs.ai cs.lg deepmind development evaluation google google deepmind practice report research responsibility safety set type