June 2026 Archive
37903.
Diffusion‑based LLMs generate many parallel tokens rather than one‑by‑one
(inceptionlabs.ai)