Unverified50% confidenceFactExact time
Continual Search continuously drives the LLM serving as judge to keep searching for unresolved diagnostic evidence through multiple rounds of iteration.
1
Sources
50%
Confidence
Long-term
Relevance
9/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFull-text search typically relies on inverted index technology to achieve millisecond-level retrieval response.70% similarUnverifiedLLMs perform reliably in high-recall screening tasks such as title-abstract screening and full-text screening, but show a marked decline in evidence extraction tasks and still require human leadership.66% similarUnverifiedSelection leakage refers to the search loop itself being able to read the final evaluation score, causing the model to overfit to the holdout set when the number of iterations is large enough, and is closely related to the multiple comparisons problem and p-hacking.66% similarUnverifiedThe ranking factors of traditional search can be partially reverse-engineered and restored through statistical analysis, but the decision-making process of large language models with billions of parameters is nearly impossible to audit at the same level of granularity.64% similarUnverified网格搜索穷举所有参数组合结果全面但耗时，随机搜索随机采样效率更高63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/921736API
curl https://kongchang.com/api/v1/knowledge/claims/921736MCP
get_claim(id=921736)