Differentially private means that:

training_algorithm(training data with a row that has "ForHackernews blood test report...") hard to distinguish from training_algorithm(training data without that) upto a factor of epsilon. They have explained further in the article itself with concrete values for epsilon.