If I read the claims correctly, it takes an unlabeled dataset + a classifier model + a handful of heuristic labeling functions, which may be both partial and in

If I read the claims correctly, it takes an unlabeled dataset + a classifier model + a handful of heuristic labeling functions, which may be both partial and inaccurate, and trains the model to be a good classifier for the dataset, without ever requiring per-example manual labeling. Quite magic.