From MoE to Looped Transformers: A Researcher Jokes About Being 8 Years Ahead

prajdabre · x · 2026-08-01

An AI researcher humorously summarized the popularization cycle of deep learning architectures. He noted that Noam Shazeer's Mixture of Experts (MoE), proposed in 2016, only became massively popular in 2024. Similarly, the Looped Transformers architecture he (and others) independently proposed in 2018 is predicted to truly explode in 2026. The post highlights the often lengthy incubation period required for cutting-edge academic research to be widely adopted by the industry.

Related event: Researcher Jokes Recurrent Transformers Will Boom in 2026(4 posts)→

Original post →

More from Fun

Fun channel →