There's no reason to believe neural nets in 2016 will bridge the gap more so than folks saying this about nets in 2000 vs in 1990 vs 1980. The introduction of GPUs, more data, and variations in the models have lead to real improvement in specific tasks, but even those get stuck again. The leap to AGI is unclear if that's what's meant. You may be referring to synthesis techniques... but same thing. (They are cool though!)