2,212 AI Papers. Zero Usable Models.

A 2021 Nature Machine Intelligence review screened 2,212 COVID-19 diagnosis and prognosis papers. 62 survived quality screening. None were judged of potential clinical use. The failures worth caring about are provenance failures: assembled datasets with the same images in train and test, and no record of where any of it came from. Full talk → https://youtu.be/v0lUudG_bwQ #kubeflow #mlops #dataengineering #machinelearning #shorts