Top New Best Ask Show Jobs

DiffusionBlocks: Block-Wise Neural Network Training | Hacker News Reader

DiffusionBlocks: Block-Wise Neural Network Training

arxiv.org

3 points·E-Reverance··5 comments

Open articleView on HN

The same Sakana that couldn't validate experiments

https://techcrunch.com/2025/02/21/sakana-walks-back-claims-t...

Other teams did a better job and provided code

https://github.com/kuleshov-group/bd3lms

This is unrelated. They both use the word "block", but what they are referring to differs

"Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models"

Yes and? The paper I linked is about network weights, not the type of generative model

Reply on news.ycombinator.com