Unverified50% confidenceOpinionExact time
General-purpose NLI models show decreased accuracy on numerical reasoning, nested negation, and long-tail entities, which is one of the core bottlenecks in audit quality.
1
Sources
50%
Confidence
Long-term
Relevance
10/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe breaking points of automated evaluation include: LLM evaluators are lenient, judge partially correct answers poorly, are sensitive to prompt wording, miss small errors, and their agreement rates can mask low recall.76% similarVerified若用含测试集的全量数据计算均值和标准差，会导致预处理泄漏，使评估指标虚高，无法真实反映模型泛化能力76% similarUnverifiedStatic analysis has a low false-positive rate but limited coverage, while LLM review has broad coverage but is unstable and expensive.72% similarUnverifiedWhen sub-agent output quality degrades and the orchestrator lacks effective anomaly detection logic, the system falls into a repeated review-failure loop and consumes API quota.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1093317API
curl https://kongchang.com/api/v1/knowledge/claims/1093317MCP
get_claim(id=1093317)