Unverified50% confidenceCautionExact time
In red-team testing, the scorer was allocated an output budget of only 50 tokens; thinking-type models exhaust all tokens on internal reasoning and ultimately return empty content.
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDuring testing, cases emerged where a model claimed 95% confidence in a move but scored 0 points according to the evaluation criteria.67% similarUnverifiedAnthropic的研究显示，即使面对毫无根据的质疑，在某些测试中模型被用户简单反驳后放弃正确答案的比例超过50%67% similarUnverifiedThe author scored only 50 points on the quiz they made themselves.67% similarUnverified置换检验通过大量随机置换（通常1000到10000次）构建零假设下的经验分布来估计p值，无需参数假设但计算成本相对较高65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/995707API
curl https://kongchang.com/api/v1/knowledge/claims/995707MCP
get_claim(id=995707)