Can foundation models label data like humans?

The article title references foundation models' capability to label data with human-level accuracy, but no article body was provided for analysis. This appears to be about AI model performance in data annotation tasks.