Diffusion‑based LLMs generate many parallel tokens rather than one‑by‑oneinceptionlabs.ai1 point·binyu··0 commentsOpen articleSaveView on HN