arXiv:2604.12376v2 Announce Type: replace Abstract: When LLM conversations grow beyond the context window, old content must be evicted -- but how does the model recover it when needed? We propose cooperative paging: evicted segments are replaced with minimal keyword bookmarks ([pN:keywords], ~8-24 tokens each), and the model is given a recall() tool to retrieve full content on demand. On the LoCoMo benchmark (10 real multi-session conversations, 300+ turns), cooperative paging achieves the highest answer quality among six methods -- outperforming truncation, BM25, word-overlap retrieval, a sea
The rapid advancement of large language models is directly confronting the limitations of existing context windows, making efficient memory management a critical and immediate bottleneck.
This research significantly enhances the practical utility and robustness of LLMs in long-duration, multi-session interactions, addressing a core challenge for complex AI applications.
LLMs can now maintain extended conversational memory more effectively, reducing 'forgetting' and enabling more sophisticated and continuous agentic behaviors.
- · LLM developers
- · AI Agent platforms
- · Enterprise AI users
- · LLMs without advanced memory solutions
- · Developers reliant on simple truncation methods
LLMs will become more capable of handling multi-turn, multi-session conversations without losing coherence or context.
This improved memory will accelerate the deployment of autonomous AI agents across various domains, as they can maintain complex long-term states.
More robust and long-context LLMs could lead to new forms of human-computer interaction and automation that were previously infeasible.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL