When enterprises need domain experts to label data or evaluate model outputs, they match and hire professionals per task through the platform (exact workflow still to be verified).
The materials do not state current alternatives; common practice would be generic crowdsourced labeling or public benchmarks (inference).
The materials only say existing AI benchmarks cover tasks poorly, without describing the concrete difficulty enterprises face; it can only be inferred that generic benchmarks fail to reflect real occupational tasks.