SOURCE-LINKED INTELLIGENCE

Separating Stream Stability from Long-Term Recall in Language Models

arXiv · AI, language, vision and robotics · article · Sep 7, 2026 · UTC

Methods for streaming language models are often discussed alongside long-context and memory systems, although they solve different problems. An attention sink can stabilize autoregressive generation over an indefinitely long stream while the model remains unable to use content that has left its recent-token cache. We argue that this distinction should be explicit in system claims and evaluation. We introduce three horizons: the stability horizon, over which predictive behavior remains well behaved; the access horizon, over which past content can still causally affect the output; and the utilit

Read original source ↗ Open in workspace

recordType

paper

region

Global