Unverified50% confidenceFactExact time
In the control condition, all judges' scores moved by 7 to 8 points, while the effect of outcome on the score was about 1 point, and it always decreased in the harm version.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedThe LLM judge serving as evaluator assigns almost all outputs a score of 7 or 8, causing it to lose discriminative power.73% similarUnverifiedIn document deskewing, even a tilt of 3–5 degrees can cause recognition rates to drop by more than 10%.68% similarUnverifiedIn the control group where the answer key was removed from the prompt, the match rate plummeted from 63% to 1% (1/75).66% similarUnverifiedRewriting one criterion in a prompt causes the pass rate of other criteria within the same prompt to fluctuate by 5 to 10 percentage points.65% similarUnverifiedAccording to the classical 1/n scaling law in statistics, quadrupling the sample size should reduce divergence by approximately 75%, but in practice divergence decreased by only 5% to 40%.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1043801API
curl https://kongchang.com/api/v1/knowledge/claims/1043801MCP
get_claim(id=1043801)