Research · Oct 25, 2026, 02:57 UTC

Reddit post introduces ASRN, a linear-memory layer for long-range copying in language models

ASRN uses bounded per-chunk attention over learned hash buckets and a learned proper-time clock to decide when to copy repeated content. The project is shared as a Zenodo preprint with a GitHub repository.

• Topicresearch