Unverified50% confidenceFactExact time
In the experiment, the proportion of unhealthy samples was 41.7%, that is, 5 out of 12 samples were unhealthy.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAmong 42 effective corruption samples, 16 were 'corrupt-and-overtrust' cases, in which the model accepted erroneous reasoning it should not have.67% similarUnverified研究表明即使是轻微的数据污染也可能导致评测分数虚高5-15个百分点67% similarUnverifiedIn the experiment, after prompts exceeded 50, the human interception rate dropped sharply from 17% to 5%, reflecting the phenomenon of approval fatigue.64% similarUnverifiedAt a significance level of ε=0.09, 63% of nodules received a definitive single-label prediction, while the remaining 37% of cases were referred to clinicians for judgment.64% similarUnverifiedIn the pilot, the LLM successfully captured 88.0% of the clinically relevant information points embedded in the cases, compared to only 38.9% for human clinicians.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1038696API
curl https://kongchang.com/api/v1/knowledge/claims/1038696MCP
get_claim(id=1038696)