A 2017 Google excerpt showed CNNs accounted for only 5% of datacenter neural network workload while ISCA 2016 papers focused almost exclusively on convolutional architectures

Positive users see the 2017 Google paper urging hardware focus beyond CNNs as proof that overlooked ideas will prove valuable for LLMs, while negative users sarcastically mock it as setting up future failures with convolutions.

2 comments with sentiment. View comments.