Unverified50% confidenceBenchmarkExact time
Fairness instructions had limited effect: family accuracy changed by only −0.3 percentage points on the cluster six-model panel and +2.5 percentage points on Kaggle, with confidence intervals spanning zero in both cases.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027
Sources
Related Entities
Related Claims
UnverifiedOn the cluster panel, replacing the model's structured decision-making with code (based on the same extracted facts) improved family accuracy by an average of 20.5 percentage points, with Llama 3.1 8B rising from 1% under self-judgment to 56% under code adjudication.69% similarUnverifiedKaggle's ranking point system decreases proportionally with team size; in a 2-person team, each member receives approximately 1/√2 of the original points.68% similarUnverifiedThe leaderboard score is the no-arbitrage rate, i.e., the proportion of families in which the bookmaker can lock in at most 5 cents; models with coverage below 90% are not scored.67% similarUnverifiedIf a difficult problem fails across all samples (pass@k ≈ 0), the reward variance is zero, the gradient signal nearly vanishes, and the parameter update approaches zero.66% similarUnverifiedAs a comparison baseline, the simple "one-prompt-per-criterion" approach achieved only kappa = 0.33 on the common subset, whereas SAGE achieved 0.58.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1003876API
curl https://kongchang.com/api/v1/knowledge/claims/1003876MCP
get_claim(id=1003876)