Unverified50% confidenceFactExact time
The Gisting method proposed by Stanford University compresses long prompts into a small number of gist tokens, AutoCompressor compresses historical context into summary vectors, and Infini-Attention introduces a compressed memory mechanism to handle infinitely long inputs.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedActive forgetting and summary compression are necessary measures to combat the unbounded growth of memory stores.71% similarVerifiedMainstream strategies for addressing short-term memory limitations include sliding window truncation, summary compression, and importance-based selective retention, with each strategy striking a different balance between information completeness and token consumption.66% similarUnverified随机上下文压缩源于大语言模型固定的上下文窗口限制，当任务历史记录超过限制时系统会自动压缩或截断早期信息66% similarUnverifiedMany sparse/compression evaluation tasks can be passed under sliding window attention, so the 5–10x compression ratios reported on these tasks are not convincing.66% similarUnverifiedMulti-head Latent Attention significantly reduces memory usage and computational overhead during inference by compressing the KV cache into a low-dimensional latent space.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/893532API
curl https://kongchang.com/api/v1/knowledge/claims/893532MCP
get_claim(id=893532)