My rule of thumb is to always have a healthy dose of skepticism regardless of how a study was published, whether in a peer-reviewed journal, preprints, or even a blog post. About bleeding edge research, publishing machine learning methods immediately on arXiv seem to be hot among academics nowadays.