Advanced Deep Learning Interview Questions #5 - The Global Accuracy Trap

We optimize for aggregate metrics, but production systems fail when slice-level performance and latency constraints are violated.

You’re in a 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?”

Most candidates say: “It must be overfitting to the validation set, or ma…