Founding and initial capabilities research - we started operations and conducted collaborations with researchers to publish our first major research result demonstrating that it is possible to cheaply undo the safety measures of an advanced open-source artificial intelligence system.