Advanced Deep Learning Interview Questions, issue 5, Mar 26, 2026
The Global Accuracy Trap
Senior ML Engineer interview at Meta, and the interviewer asks:
“You just bumped a model’s accuracy from 75% to 85%, crossing the business cutoff for deployment. In what scenario does deploying this mathematically ‘better’ model actually destroy the end-user experience?”
Don’t say: “It must be overfitting to the validation set, or maybe there’s data leakage!”
We optimize for aggregate metrics, but production systems fail when slice-level performance and latency constraints are violated.
The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.