The source of this stuff is PubPeer, a website where people can comment on papers and do post-publication peer review. A quick rummage through this site is the quickest way to destroy your confidence in academic science for good. There's no direct equivalent of Photoshopped Western Blots for computer science, but if you want a laugh search for papers generated by SciGen, a perl script that makes joke papers:<p><a href="https://pubpeer.com/search?q=scigen" rel="nofollow">https://pubpeer.com/search?q=scigen</a><p>Or for a more serious example of problems (medical AI):<p><a href="https://pubpeer.com/publications/ABB8F161E6691210C44545FF8990C1" rel="nofollow">https://pubpeer.com/publications/ABB8F161E6691210C44545FF899...</a><p><i>The authors report an AUROC of 1.00. This is extremely high and implies perfect discrimination between cases and controls, which seems unlikely given the highly heterogeneous nature of ASD and suggests some sort of data leakage<p>...<p>The highlighted point estimates are not in the middle of their confidence intervals, either by a modest amount (yellow) or by a large amount (pink). In some pink cases, the point estimate is not inside the CI at all.<p>...<p>I have rather strong concerns regarding the apparent absence of a hold-out dataset.</i>