LGCLFeb 5, 2025

Do Large Language Model Benchmarks Test Reliability?

arXiv:2502.03461v143 citationsh-index: 8Has Code
Originality Incremental advance
AI Analysis

This addresses the gap in evaluating reliability for large language models, which is crucial for safe deployment, though it is incremental as it builds on existing benchmarks.

The study found that current large language model benchmarks are compromised by pervasive label errors, which obscure model failures and unreliable behavior, and proposed platinum benchmarks that reveal frontier models still struggle on simple tasks like elementary math word problems.

When deploying large language models (LLMs), it is important to ensure that these models are not only capable, but also reliable. Many benchmarks have been created to track LLMs' growing capabilities, however there has been no similar focus on measuring their reliability. To understand the potential ramifications of this gap, we investigate how well current benchmarks quantify model reliability. We find that pervasive label errors can compromise these evaluations, obscuring lingering model failures and hiding unreliable behavior. Motivated by this gap in the evaluation of reliability, we then propose the concept of so-called platinum benchmarks, i.e., benchmarks carefully curated to minimize label errors and ambiguity. As a first attempt at constructing such benchmarks, we revise examples from fifteen existing popular benchmarks. We evaluate a wide range of models on these platinum benchmarks and find that, indeed, frontier LLMs still exhibit failures on simple tasks such as elementary-level math word problems. Analyzing these failures further reveals previously unidentified patterns of problems on which frontier models consistently struggle. We provide code at https://github.com/MadryLab/platinum-benchmarks

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes