factual

Adversarial examples show deep networks use non-human statistical shortcuts

Moitra argues that deep networks solve object recognition through statistical relationships humans do not use, shown by adversarial patterns that fool detectors and by experiments where networks trained on mislabeled adversarial data still generalize to natural data, so AI success is not human-like.

factualpending

Speaker

Ankur Moitra

Evidence Quote

“when you train a deep network on this with all the wrong labels, it's gonna learn something that does well on natural data”

Source

Lex Fridman Podcast #503 - AI Crisis in Math— Lex Fridman Podcast
Created: 11/17/2026, 6:25:32 AM

Your notes

Loading notes...