@varx@infosec.exchange This doesn't quite do what you're going for, I think, but I've found "the bias bias" to be useful so I thought I'd share. It was coined by Henry Brighton in a 2015 paper and refers to the tendency to reach for increasingly complicated models of the world in all sorts of fields. It is opposed to seeking simpler models that might work better, fit the data better, be easier to understand and reason about, etc. Brighton's work on this, and the field of ecological rationality generally, uncovers examples where simpler is better as well as criteria for deciding when you should reach for complex models and when doing so won't help or will produce worse results.

In the present context, always reaching for more training data and models with more tunable weights is an example of this bias in action, seems to me.