Been reading biomedical study proposals. It is remarkable how much old-fashioned concepts such as study design, sample size, biostats, "is there a plausible path from the data we'll obtain to the question we state?", have given way to "we'll collect multiomics and AI will figure something out"