Unverified50% confidenceBenchmarkExact time
Across 72 matchups, the four models (gemma2, qwen2.5, llama3.1, mistral) had yield rates of 50%, 56%, 61%, and 78%, respectively.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/13/2026
First Seen
Valid until: 1/11/2027
Sources
Related Entities
Related Claims
UnverifiedIn the specific experimental case, the overlapping-anchor scheme (n=912) showed the model at 55.2% versus the majority class at 51.1%, a net advantage of +4.1pp, about 2.4 standard errors; the non-overlapping-anchor scheme (n=163) showed 59.5% versus 54.0%, a net advantage of +5.5pp, about 1.4 standard errors.66% 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.65% similarUnverifiedThe framework's prediction intervals achieve empirical coverage rates of 0.533, 0.831, and 0.958 at nominal confidence levels of 50%, 80%, and 95%, respectively.64% similarUnverifiedOn 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%.64% similarUnverifiedAt the mention level, the model achieved an overall precision of 74.1% and a recall of 70.5%.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1019897API
curl https://kongchang.com/api/v1/knowledge/claims/1019897MCP
get_claim(id=1019897)