Which ML subfields are becoming irrelevant?

A Reddit post questions whether subfields like neural architecture search (3,000+ models in five years), adversarial ML (9,000 papers with no practical use) and fairness research deserve continued investment, urging a rational reassessment.

• Study: Pretraining Poisoning Can Plant Agentic Backdoors2026-11-02