AnalysisAI ModelsOctober 2, 2026

Cohere Labs talk covers discrete diffusion speedups for LLMs

Read original source →youtube.com

Subham Sahoo, a senior research scientist at MBZUAI-IFM, presents on lossless speedups for LLMs via discrete diffusion, arguing autoregressive models stay slow due to sequential decoding. Hosted by Cohere Labs research engineer Saurabh Dash.

People · Subham Sahoo, Saurabh Dash

1 source

More stories today

Open the live feed