[mturmon]

The way it's worded is not 100% clear. Hinton, who is an excellent lecturer and explainer, is talking about neural nets trained with "deep learning" techniques (not vanilla single-hidden-layer nets), which have had striking success at hard vision problems that have been difficult to solve top-to-bottom with SVMs (e.g., you could get good performance from an SVM, but you'd have to go on a hunt for good low-level features first).

That said, there is a rather unhelpful herd mentality in the field, with people moving from one Next Big Thing to another, disparaging the previous Big Thing along the way.