AI Models Tested for Drone Surveillance Capabilities
Anthropic and Andon Labs have developed a benchmark, Drone-Bench, to evaluate AI models' ability to autonomously control drones for surveillance tasks. The project aims to assess models' capabilities in locating and following targets using drones, highlighting the dual-use nature of AI technology. The evaluation involves decomposing the task into sub-tasks, such as navigation and detection, to measure models' performance against a human-AI team baseline. The findings indicate progress in AI models' ability to handle complex surveillance tasks, with implications for both civilian and military applications. 
