Unverified50% confidenceFactExact time
The researchers drew 60 decisions from the late stage of each model's matches and, when the records were sufficient, compared the changes in probability distributions under three conditions: the original question, a repeated question, and removing the records.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAnswer balancing requires that the yes/no labels of each question type be roughly matched in distribution, so that the expected accuracy of random guessing stays around 50%.74% similarUnverifiedRanked by question difficulty, persistence rate was 91.4% on the 60 questions that all models answered correctly, but only 73.5% on the 52 questions that at least one model got wrong, indicating that sycophancy tracks uncertainty.71% similarUnverified一个随机猜测的模型在单次二元预测中有50%的概率押中71% similarUnverifiedThe experiment has a small sample size (single model, single test, 50 questions), is subject to random fluctuation, and constitutes a heuristic finding rather than a definitive conclusion.70% similarUnverifiedA rule was set in the experiment: any judgment with a confidence level below 60% would be directly handed over to human processing.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1047725API
curl https://kongchang.com/api/v1/knowledge/claims/1047725MCP
get_claim(id=1047725)