LAW: A language model built purely on self-conditioning

Researchers introduced LAW (Loopholing All the Way), a language model that operates purely on self-conditioning instead of diffusion sampling, extending prior Loopholing Discrete Diffusion work and challenging recent efforts to remove self-conditioning from diffusion language models.

2026-10-22 ~ 2026-10-23 · 3 related posts

• LAW: a language model built purely on self-conditioning — caglarml · 2026-10-22
• LAW proposes dropping diffusion entirely: language models built on pure self-conditioning — ArnaudDoucet1 · 2026-10-23
• LAW: a language model built purely on self-conditioning, extending Loopholing discrete diffusion — caglarml · 2026-10-23