bias-variance tradeoff

Are there any theoretically-guided practices left in machine learning nowadays? [D]

A Reddit forum post questions whether theoretical foundations still meaningfully guide machine learning practice, noting that long-held pedagogical principles (e.g., overfitting from too much data, optimizer selection by convergence guarantees) have been empirically violated at scale without performance loss.