New benchmark exposes how badly AI struggles with real knowledge work

A new benchmark reveals that leading AI models can only fully solve 3% of realistic knowledge work tasks. This significant shortfall highlights the limitations of current AI capabilities in practical applications, emphasizing the need for improvements in model architectures and training methodologies for practitioners developing AI solutions.