Here is another heuristic: if the paper is published in Nature/Science/Cell (for bio-related applications), then highly likely to be wrong/not generalizable.