Explore key challenges in building and evaluating trustworthy AI systems through this 80-minute lecture that begins with a foundational recap of human trust in AI. Learn what factors influence people's trust in reliable models before diving deep into critical issues in AI evaluation, including data shortcuts in benchmarking, the importance of independent test instances, maintaining basic system functionality, and ensuring robust evaluation methods. Examine the complexities of securing high-quality pretraining data and understand how these challenges impact the development of dependable AI systems that can earn justified user trust.