LLMEval-3: A Large-Scale Longitudinal Study on Robust and Fair Evaluation of Large Language Models
This addresses the critical issue of unreliable evaluation for LLM developers and researchers, offering a more robust methodology, though it is incremental in improving existing evaluation paradigms.
The paper tackles the problem of data contamination and leaderboard overfitting in LLM evaluation by introducing LLMEval-3, a dynamic evaluation framework that uses a proprietary bank of 220k questions and an automated pipeline, resulting in a 20-month study revealing performance ceilings and vulnerabilities in nearly 50 models.
Existing evaluation of Large Language Models (LLMs) on static benchmarks is vulnerable to data contamination and leaderboard overfitting, critical issues that obscure true model capabilities. To address this, we introduce LLMEval-3, a framework for dynamic evaluation of LLMs. LLMEval-3 is built on a proprietary bank of 220k graduate-level questions, from which it dynamically samples unseen test sets for each evaluation run. Its automated pipeline ensures integrity via contamination-resistant data curation, a novel anti-cheating architecture, and a calibrated LLM-as-a-judge process achieving 90% agreement with human experts, complemented by a relative ranking system for fair comparison. An 20-month longitudinal study of nearly 50 leading models reveals a performance ceiling on knowledge memorization and exposes data contamination vulnerabilities undetectable by static benchmarks. The framework demonstrates exceptional robustness in ranking stability and consistency, providing strong empirical validation for the dynamic evaluation paradigm. LLMEval-3 offers a robust and credible methodology for assessing the true capabilities of LLMs beyond leaderboard scores, promoting the development of more trustworthy evaluation standards.