Unverified50% confidenceBenchmarkExact time
The LoRA baseline used for comparison in the paper uses rank 19 with 6.6 million parameters, achieving a correction rate of 83.3%, but exhibited capability loss of 30% to 75%.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2023年的分析显示主流LLM在生成学术引用时的错误率从30%到超过60%不等67% similarUnverified2012年Hinton团队的AlexNet以top-5错误率从26%骤降至15%的成绩震惊学界64% similarUnverifiedThe standard margin of error for Gallup polls is typically kept within ±3–4 percentage points at a 95% confidence level.64% similarUnverifiedThe DeepSuite benchmark has a false positive rate of 44.9% and a false negative rate of 1.2%.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924668API
curl https://kongchang.com/api/v1/knowledge/claims/924668MCP
get_claim(id=924668)