hkust

#hkust

Less Uniform Discrete Diffusion is More Powerful and Scalable

arXiv cs.CL ↗ · 2d ago Cached

This paper proposes LUDI, a less uniform diffusion language modeling framework that fixes over-uniform training objectives and condition-target confusion in uniform diffusion LMs, enabling a 7B-scale UDLM with 3x-token-per-step speedup over autoregressive decoding and competitive complex reasoning performance.

0 favorites 0 likes

#hkust

@vintcessun: Frustrated staring at a blank document late at night? Finished your paper but feeling unsure? Advisor too busy, no one to evaluate your ideas. The problem isn't that you're not working hard enough; you need a research co-supervisor always at hand. A HKUST professor distilled his ten years of publishing and reviewing experience into an AI-executable skill set. From evaluating ideas, generating introduction outlines…

X AI KOLs Timeline ↗ · 2026-07-09 Cached

HKUST Assistant Professor Luo Yuyu released the open-source project Supervisor-Skills, which distills ten years of research experience into AI skills executable by large models, covering topic selection, writing, reviewing, etc., aiming to become an AI co-supervisor for researchers.

0 favorites 0 likes

hkust

Less Uniform Discrete Diffusion is More Powerful and Scalable

Submit Feedback