Mihesche

[Mihesche @mihesche]

🔗 Code: https://t.co/wN3sSciJIA

[Mihesche @mihesche]

Why are we still generating language one token at a time? 🧵

Autoregressive models like GPT dominate, but they force us into a sequential bottleneck.

Diffusion Language Models (DLMs) offer a parallel alternative—but they have a hidden efficiency gap.

[Mihesche @mihesche]

Read more:
🔗 Paper: https://t.co/Cm5Ulx8mdk

[Mihesche @mihesche]

Exciting news: our work on NVSA just appeared in @Nature Machine Intelligence.

Many thanks to the amazing team at @IBMResearch and @ETH_en:
@13_musti
@LucaBeniniZhFe
Abu Sebastian
Abbas Rahimi

Paper: https://t.co/ehUw0GJFq8
https://t.co/TRn7pg9isT