ByteBrief
We're a portrait publication through and through. Turn your phone back and your briefing picks up right where you left it.
(We tried widescreen once. It wasn't us.)
Researchers introduce the Lattice of Latents, a language model architecture that increases computational capacity via horizontal latent streams. On 20B FineWeb-Edu tokens, a 406M-parameter Lattice model lowers validation perplexity from 15.41 to 14.40 over a parameter-matched Transformer. With hyper-connections and input residuals, perplexity drops further to 13.61.
Tap to vote and see what everyone thinks.
Summary by ByteBrief