logoalt Hacker News

_0ffhyesterday at 9:14 AM0 repliesview on HN

Good news! On the current trajectory, I am very hopeful nonlinear RNNs could make a comeback. Which would incidentally help to ease the memory pressure for inference tasks.