AI & ML
SOTA or Luck? The Winner's Curse in LLM Leaderboards
arXiv stat.ML0 views
arXiv:2611.22227v1 Announce Type: cross Abstract: Public LLM leaderboards guide model evaluation and adoption, yet the reported performance of top-ranked models can be systematically overstated. This leaderboard winner's curse arises because ranking noisy benchmark scores rewards sampling luck alongside genuine ability. The same selection effect can distort scores throughout the leaderboard, a phenomenon we call the rank-induced bias. We develop an $m$-out-of-$N$ bootstrap to correct for this bias. It assumes a fixed model pool with independent, identically distributed bounded tasks while allowing arbitrary within-task correlation among models. Under exact population tiers separated by positive gaps, we prove that the bootstrap reduces $\Theta(N^{-1/2})$ ranking bias to expected residual bias $o(N^{-1/2})$. Across 14 benchmarks from 5 benchmark families, the rank-1 adjustments range from approximately 0.09 to 3.86 percentage points. In our SWE-bench Verified snapshot, the tied leaders receive an adjustment of 0.90 percentage points. Such overstatement is large enough to make the apparent winner statistically fragile. Thus, we use the same bootstrap to further quantify the intervals of plausible rank positions for each model and group models into tiers. We establish sufficient conditions for consistent recovery of population tiers. On SWE-bench Verified, this procedure places 13 of 173 models in the leading tier. We combine the score adjustment and rank intervals in the Rank Uncertainty and Leaderboard Evaluation Report (RULER). We validate its finite-sample performance using sample splitting and calibrated simulations, and advocate publishing RULER alongside raw rankings so readers can assess both reported performance and the precision of the ordering. RULER also provides guidance for benchmark design that accounts for selection effects and ranking uncertainty.
Source:https://arxiv.org/abs/2611.22227Related posts
Gemini price increases: Carbon and Argon double on 1 December 2026, Flash on 1 January 2027
0dev.to
OpenAI shuts down whisper-1 and its other transcription models on 26 February 2027: a migration checklist
0dev.to
YouTube Partner Program changes on 1 February 2027: the 31 January deadline and the new entry thresholds
0dev.to
The ‘GTA 6’ Work Blackout Is Here
0Wired
Best Buy sells Blu-rays online and in-store again
0Ars Technica

Sony’s impressive Bravia 8 II OLED hits a new low price ahead of Black Friday
0The Verge