Unverified50% confidenceBenchmarkExact time
The calibration method proposed in "Found in the Middle" weakens attention bias and improves RAG results by up to 10 percentage points over existing methods.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedUnder fixed compute, single-head attention scores 0.9 points lower than the optimal setting, and quality actually decreases when there are too many heads.73% similarUnverifiedIn document deskewing, even a tilt of 3–5 degrees can cause recognition rates to drop by more than 10%.70% similarUnverifiedMeta、NVIDIA等公司的研究表明RAG在知识密集型任务上的准确率比纯参数记忆提高20%-40%68% similarUnverified在HotpotQA等多跳推理基准上，ReAct相比Act-only方案的任务完成率提升约10-15个百分点，同时幻觉率显著降低66% similarVerifiedMLA can reduce the KV Cache VRAM usage to 5%–13% of traditional multi-head attention.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1004542API
curl https://kongchang.com/api/v1/knowledge/claims/1004542MCP
get_claim(id=1004542)