I'm in total agreement with Marcus on this one. The authors' misapplied statistics notwithstanding, I am a qualified expert on designing biologic assays, and I have used GPT-4 very successfully to help design a couple assays. My experience has been that it is a really good crutch for that sort of work, insofar as it can point you in the right direction. It can't point you to papers, but it can help you refine search terms which can then be brought to PubMed or another database as a refiner. All that said, the things that GPT has helped me with are things I could have done without it, but it sped the process up dramatically. It stands to reason that similar would be true for any arbitrary assay design.

As for the abused stats in the paper, my guess is that it was intentional, and that Marcus is giving them too much benefit of the doubt by saying it was probably someone who doesn't know what they're doing. If it were biologists I'd buy it. But these are math nerds who likely know very well what a Bonferroni correction is and how it's meant to be used. Bonferroni is one of the strictest post hoc correction methods. It's meant to weed out all but the most robust results. There are other less severe methods that are just as common, so the choice to use a strict methods has the air of deceitfulness. As in, it would make a lot of sense to use the strictest statistical methods if what you were doing was trying to prove a point and not just following the data.