Unverified50% confidenceFactExact time
The paper introduces pre-window curvature c_m, which quantifies the degree to which states visited more than m steps earlier reduce the current marginal gain.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedA smaller c_m indicates weaker historical influence, allowing the policy to be more myopic; a larger c_m requires retaining a longer history window.64% similarUnverifiedOn reference-based metrics, recent LLMs have greatly narrowed the gap with fine-tuned models, even matching them in some settings.60% similarUnverifiedStandard attention's computation grows quadratically with sequence length, whereas the low-rank bottleneck decomposes the transformation into two steps: first compressing to r dimensions and then expanding.59% similarUnverifiedThe paper proves that the SGD iterates converge to a stable OU process at the central fluctuation scale and uses this limit to characterize the tail probabilities.58% similarUnverified研究表明当上下文窗口填满时，LLM对序列首尾信息的关注度显著高于中间部分，即'迷失在中间'效应58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134352API
curl https://kongchang.com/api/v1/knowledge/claims/1134352MCP
get_claim(id=1134352)