REVIEW 19 cited by
Benchmarking Benchmark Leakage in Large Language Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Amid the expanding use of pre-training data, the phenomenon of benchmark dataset leakage has become increasingly prominent, exacerbated by opaque training processes and the often undisclosed inclusion of supervised data in contemporary Large Language Models (LLMs). This issue skews benchmark effectiveness and fosters potentially unfair comparisons, impeding the field's healthy development. To address this, we introduce a detection pipeline utilizing Perplexity and N-gram accuracy, two simple and scalable metrics that gauge a model's prediction precision on benchmark, to identify potential data leakages. By analyzing 31 LLMs under the context of mathematical reasoning, we reveal substantial instances of training even test set misuse, resulting in potentially unfair comparisons. These findings prompt us to offer several recommendations regarding model documentation, benchmark setup, and future evaluations. Notably, we propose the "Benchmark Transparency Card" to encourage clear documentation of benchmark utilization, promoting transparency and healthy developments of LLMs. we have made our leaderboard, pipeline implementation, and model predictions publicly available, fostering future research.
Forward citations
Cited by 19 Pith papers
-
Navigating the Open-Source Model Ecosystem: An Empirical Study of Creator Practices in Artistic Image Generation
A novel 6M-image Pixiv dataset shows open-source image generation has long-tail model usage, slow life cycles with version inertia, and surging multi-LoRA customization linked to higher engagement.
-
Can Vision Language Models Understand Mimed Actions?
Vision-language models identify real actions with context far better than they identify mimed actions performed by 3D avatars, while humans are equally accurate on both.
-
SenWorld: A Digital-Twin Simulation for Generating Context-Rich Evaluation Data
SenWorld simulates a fake smartphone-and-person world to generate assistant test cases whose correct answers are fixed pointers to existing records.
-
Beyond Success Rate: Cost-Aware Evaluation of Offensive and Defensive Security Agents
Security-agent success changes differently with budget: offensive CTF tasks improve with more compute, while defensive SOC work depends more on tool discipline than spend.
-
PITMuS: A Tool for Automated Bug Dataset Generation via Source-Level Mutant Reconstruction
PITMuS reconstructs source-level original/mutated method pairs from PIT bytecode mutation reports at 99.96% coverage across eight Java projects.
-
SoftMatcha 2: A Fast and Soft Pattern Matcher for Trillion-Scale Corpora
Suffix-array exact lookup plus word-vector similarity enables semantic search over 1.4T tokens at 278 ms p95, 33x faster exact lookup than infini-gram, and flags benchmark contamination exact matchers miss.
-
FiMMIA: scaling semantic perturbation-based membership inference across modalities
A perturbation-based membership inference framework for multimodal LLMs reports high AUC-ROC on fine-tuned image/video/audio models and shows that several existing MIA benchmarks are confounded by distribution shifts.
-
MatSciBench: Benchmarking the Reasoning Ability of Large Language Models in Materials Science
MatSciBench, a 1,340-question materials science benchmark with detailed solutions and images, shows top LLMs still fall short of college-level mastery.
-
Exploiting Leaderboards for Large-Scale Distribution of Malicious Models
A new attack framework, TrojanClimb, shows that adversaries can place models with embedded backdoors or biases on public leaderboards while retaining competitive rankings, across text embeddings, text generation, spee...
-
Deprecating Benchmarks: Criteria and Framework
A framework for deprecating outdated or flawed AI benchmarks, with seven criteria and a three-phase process of assessment, reporting, and notification.
-
ConsistencyChecker: Tree-based Evaluation of LLM Generalization Capabilities
ConsistencyChecker ranks LLMs by how well they survive chains of reversible transformations, and those scores track WMT 2024 translation quality rankings (r > 0.7) without using WMT paired data.
-
OIBench: Benchmarking Strong Reasoning Models with Olympiad in Informatics
A private, contamination-resistant benchmark of 250 olympiad-level programming problems shows top reasoning models reaching about 36% solve rates, far above conventional models.
-
Chain of Methodologies: Scaling Test Time Computation without Training
A prompt framework that interleaves methodology selection with reasoning steps improves LLM math and QA accuracy when combined with a Python interpreter, but the benefit of the methodology selection itself is small an...
-
Establishing Trustworthy LLM Evaluation via Shortcut Neuron Analysis
Shortcut neuron patching suppresses benchmark-contamination shortcuts in LLMs and yields evaluation scores that strongly correlate with the external MixEval benchmark.
-
Simulating Training Data Leakage in Multiple-Choice Benchmarks for LLM Evaluation
Under simulated leakage, n-gram-based detection beats permutation and truncation methods, and cleaning flag-prone MMLU instances changes model rankings only slightly.
-
Evaluating the Sensitivity of LLMs to Prior Context
Prior conversational context, especially from a different knowledge domain, can sharply reduce LLM multiple-choice accuracy, and repeating the task near the query mitigates the drop.
-
Stochastic Chameleons: Irrelevant Context Hallucinations Reveal Class-Based (Mis)Generalization in LLMs
LLMs systematically combine abstract category cues from a query with features from irrelevant context, causing structured answer flips, a behavior the authors call class-based (mis)generalization.
-
LemmaBench: A Live, Research-Level Benchmark to Evaluate LLM Capabilities in Mathematics
A live benchmark auto-extracts self-contained lemmas from recent arXiv papers and finds top LLMs solve only 10–15% at pass@1.
-
Correcting Prompt Dependence in LLM Benchmarks: A Bayesian Hierarchical Model with Embedding-Space Clustering
A Bayesian model that groups similar LLM test prompts into clusters gives better predictive scores than a no-clustering baseline but does not prove that it truly corrects prompt dependence.
Discussion (0). Continue with ORCID to comment.