Unverified50% confidenceBenchmarkExact time
On Qwen3-4B, the behavioral test yielded a faithfulness score of 88.2%, which is 11.4 percentage points higher than the causally measured 76.9%.
1
Sources
50%
Confidence
Long-term
Relevance
9/24/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在歧义点的四类行为中，'先搜再问'成功率最高，平均通过率达93.4%，远高于'直接猜'的56.5%，'搜多次再猜'为51.5%69% similarUnverifiedOn the Qwen3-4B model, 76.9% (±2.8%) of stated reasoning steps at the most sensitive middle layers of the network were causally load-bearing.67% similarUnverified研究表明人类评估者在判断LLM输出的事实准确性时，正确率往往只有50%-70%67% similarUnverified对同一测试用例执行多次调用统计成功比例得到通过率，比二元的通过/失败判断更能反映模型真实表现66% similarUnverifiedThe gap between behavioral and causal tests is statistically highly significant (item-matched pairing, 111:14 discordant pairs, p < 1e-15).65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/946173API
curl https://kongchang.com/api/v1/knowledge/claims/946173MCP
get_claim(id=946173)