Advances in AI and machine learning, and specifically large language models such as ChatGPT, Bard and Claude, herald a transformational period in access to knowledge.   This grant provides three years of support to the Wikimedia Foundation, the parent organization of Wikipedia, for a new initiative to harness these advances to improve Wikipedia’s core function—access to reliable knowledge for half a billion people each month—and to protect the largest encyclopedia in human history from vandalism and even potential obsolescence.  Planned activities over the grant period include the iterative improvement and user testing of a Wikipedia plug-in using ChatGPt—now available in an experimental beta version and generating 1000 queries per day; use of AI and machine learning tools to suggest valuable edits that human editors could use to improve articles; an improved open-source neural machine translation model that supports over 200 languages, significantly boosting Wikipedia’s translation capabilities; machine learning tools to help moderators filter bad edits; a machine learning model that will let volunteers build and host their own AI models and tools (versus those built by Wikipedia staff); and a new ML moderator tool called Automoderator that will defend against vandalism with automated prevention or reversion of bad edits.  Wikipedia staff will also experiment and observe the evolving ways that contributors, volunteers, donors, and users interact use AI tools to interact with the site and whether infrastructure changes are needed to accommodate chatbot use or other tools.