That's fair pushback re how "standard" deep learning is. That said, those methods are rapidly establishing themselves as the go-to for applications in speech, vision, and text. At least for those outfits that can afford the substantial dev costs.