[andy99]

You've probably seen it, but a more recent, related paper (that I think has some of the same authors) about inherent features of modern ML that make models so fragile, even if they test OK:

Underspecification Presents Challenges for Credibility in Modern Machine Learning, D'Amour et al., https://arxiv.org/abs/2011.03395

[kendallpark]

I had not seen it yet, excited to read it! Thanks!