LLM Olympiad: Why Model Evaluation Needs a Sealed Exam

TL;DR

The paper advocates for an Olympiad-style evaluation approach for LLMs, where sealed problems and standardized testing improve trustworthiness and transparency of model assessments.

Contribution

It proposes a novel evaluation framework combining sealed problems, pre-submitted entries, and post-evaluation transparency to enhance trust in LLM performance assessments.

Findings

Reduces risks of score manipulation and hidden evaluation choices.

Enhances reproducibility and auditability of results.

Promotes fair comparison through standardized testing.

Abstract

Benchmarks and leaderboards are how NLP most often communicates progress, but in the LLM era they are increasingly easy to misread. Scores can reflect benchmark-chasing, hidden evaluation choices, or accidental exposure to test content -- not just broad capability. Closed benchmarks delay some of these issues, but reduce transparency and make it harder for the community to learn from results. We argue for a complementary practice: an Olympiad-style evaluation event where problems are sealed until evaluation, submissions are frozen in advance, and all entries run through one standardized harness. After scoring, the full task set and evaluation code are released so results can be reproduced and audited. This design aims to make strong performance harder to ``manufacture'' and easier to trust.