Unverified50% confidenceSolutionExact time
LLMs perform reliably in high-recall screening tasks such as title-abstract screening and full-text screening, making them suitable for independent or semi-automated processing, while evidence extraction tasks still require human leadership.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/11/2026
First Seen
Valid until: 12/10/2026
Sources
Related Entities
Related Claims
Unverified对于需要审查完整轨迹的复杂Skill，可以引入LLM as a Judge配合评分标准（rubric）来判断通过或失败72% similarUnverifiedLLM评估需同时考量事实准确性、指令遵循度、有害性、有用性和表达质量等多个维度，这些维度之间常存在张力70% similarUnverified在缺乏 review 的环境下，可通过为常见任务建立质量检查清单（数据质量核验、口径一致性、结论与数据匹配等）实现系统化自我审查70% similarUnverified标准化基准（如MMLU、HumanEval、GSM8K）任务边界清晰、上下文适中，与真实工程调试场景存在分布偏移70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/895597API
curl https://kongchang.com/api/v1/knowledge/claims/895597MCP
get_claim(id=895597)