REVIEW 32 cited by
Rethinking Benchmark and Contamination for Language Models with Rephrased Samples
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
Rethinking Benchmark and Contamination for Language Models with Rephrased Samples
read the original abstract
Large language models are increasingly trained on all the data ever produced by humans. Many have raised concerns about the trustworthiness of public benchmarks due to potential contamination in pre-training or fine-tuning datasets. While most data decontamination efforts apply string matching (e.g., n-gram overlap) to remove benchmark data, we show that these methods are insufficient, and simple variations of test data (e.g., paraphrasing, translation) can easily bypass these decontamination measures. Furthermore, we demonstrate that if such variation of test data is not eliminated, a 13B model can easily overfit a test benchmark and achieve drastically high performance, on par with GPT-4. We validate such observations in widely used benchmarks such as MMLU, GSK8k, and HumanEval. To address this growing risk, we propose a stronger LLM-based decontamination method and apply it to widely used pre-training and fine-tuning datasets, revealing significant previously unknown test overlap. For example, in pre-training sets such as RedPajama-Data-1T and StarCoder-Data, we identified that 8-18\% of the HumanEval benchmark overlaps. Interestingly, we also find such contamination in synthetic dataset generated by GPT-3.5/4, suggesting a potential risk of unintentional contamination. We urge the community to adopt stronger decontamination approaches when using public benchmarks. Moreover, we call for the community to actively develop fresh one-time exams to evaluate models accurately. Our decontamination tool is publicly available at https://github.com/lm-sys/llm-decontaminator.
Forward citations
Cited by 32 Pith papers
-
TSFMAudit: Data Contamination Auditing in Forecasting Time Series Foundation Models
TSFMAudit detects pretraining contamination in time series foundation models via probe adaptation dynamics (faster loss drop, smaller backbone shift), tested on 6 models and 187 datasets against 10 LLM-derived baselines.
-
Which Models Are Our Models Built On? Auditing Invisible Dependencies in Modern LLMs
ModSleuth reconstructs dependency graphs from public artifacts for four LLM releases, recovering 1,060 source-verified dependencies and exposing license issues, train-evaluation coupling, and documentation gaps.
-
How Hard is it to Rig a Benchmark? A Social Choice Analysis of Leaderboard Robustness
Benchmark-specific training maps to shift bribery and is NP-hard under Borda and mean win rate; mean win rate has the highest instance-level robustness (median 22 tasks on BBH) among tested aggregation rules.
-
Provable Joint Decontamination for Benchmarking Multiple Large Language Models
JECS aggregates per-model conformal p-values via their maximum and reconstructs a conservative envelope of the max-p null distribution to select benchmarks with global contamination rate control.
-
Do Androids Dream of Breaking the Game? Systematically Auditing AI Agent Benchmarks with BenchJack
BenchJack audits 10 AI agent benchmarks, synthesizes exploits achieving near-perfect scores without task completion, surfaces 219 flaws, and reduces hackable-task ratios to under 10% on four benchmarks via iterative patching.
-
Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity
A new paired-prompt protocol reveals alignment-pipeline-specific heterogeneity in how open-weight LLMs respond to evaluation versus deployment framings.
-
Agent Island: A Saturation- and Contamination-Resistant Benchmark from Multiagent Games
Agent Island is a new multiagent game environment that functions as a dynamic benchmark resistant to saturation and contamination, with Bayesian ranking showing OpenAI GPT-5.5 as the strongest performer among 49 model...
-
When Prompt Under-Specification Improves Code Correctness: An Exploratory Study of Prompt Wording and Structure Effects on LLM-Based Code Generation
Structurally rich task descriptions make LLMs robust to prompt under-specification, and under-specification can enhance code correctness by disrupting misleading lexical or structural cues.
-
DRBENCHER: Can Your Agent Identify the Entity, Retrieve Its Properties and Do the Math?
DRBENCHER generates multi-hop questions across biochemistry, finance, geophysics, security, and history that test interleaved browsing and computation, where the strongest models reach only 20% accuracy and human vali...
-
LogitTrace: Detecting Benchmark Contamination via Layerwise Logit Trajectories
LogitTrace detects benchmark contamination by showing that contaminated inputs produce earlier stabilization in layerwise logit trajectories while clean inputs show more gradual accumulation.
-
Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI
A post-hoc audit framework finds that many agent benchmark scores are inflated by protocol exposures (public answers, readable hidden state, generator regularities, feedback, or scoring flaws), with measured inflation...
-
SrDetection: A Self-Referential Framework for Data Leakage Detection in Code Large Language Models
SrDetection detects data leakage in Code LLMs via contrast between original benchmark samples and their semantic variants, reporting F1 gains of 21.52 (gray-box) and 14.46 (black-box) over baselines in a controlled testbed.
-
Embodied-BenchClaw: An Autonomous Multi-Agent System for Embodied Spatial Intelligence Benchmark Construction
Embodied-BenchClaw deploys three coordinated agents and a reusable Skill Library to automatically generate verifiable embodied spatial benchmarks across indoor/outdoor reasoning, manipulation, navigation, and aerial t...
-
Towards Evaluation Engineering: An Empirical Study of ML Evaluation Harnesses in the Wild
An empirical study of 57 ML evaluation harnesses shows 41.4% of operational issues occur in the specification stage, driven mainly by unimplemented features, documentation gaps, and missing input validation.
-
TRACER: A Semantic-Aware Framework for Fine-Grained Contamination Detection in Code LLMs
TRACER presents a semantic-aware framework and the first benchmark for fine-grained code contamination detection across three levels of overlap, reporting F1 scores of 0.91-0.92 and large gains over prior methods.
-
Decaf: Improving Neural Decompilation with Automatic Feedback and Search
Decaf uses compiler feedback and search to improve neural decompilation, boosting semantic success rate from 26.0% to 83.9% on ExeBench Real -O2 split.
-
Can Agent Benchmarks Support Their Scores? Evidence-Supported Bounds for Interactive-Agent Evaluation
Agent benchmarks can report evidence-supported score bounds instead of single misleading success rates by adding a layer that checks required artifacts for outcome verification.
-
GSM-SEM: Benchmark and Framework for Generating Semantically Variant Augmentations
GSM-SEM is a reusable framework for creating semantically variant augmentations of math benchmarks like GSM8K that alter facts but preserve answers and difficulty, with evaluations showing LLM performance drops of up ...
-
GSM-SEM: Benchmark and Framework for Generating Semantically Variant Augmentations
GSM-SEM generates reusable, stochastic semantic variants of math reasoning benchmarks that alter underlying facts but preserve answers, producing larger LLM performance drops than prior surface-level variants.
-
TPS-CalcBench: A Benchmark and Diagnostic Evaluation Framework for LLM Analytical Calculation Competence in Hypersonic Thermal Protection System Engineering
TPS-CalcBench is a new benchmark and evaluation framework that tests LLMs on analytical calculations in hypersonic aerodynamics and gas dynamics, using dual-track scoring and interventions to detect physically invalid...
-
League of LLMs: A Benchmark-Free Paradigm for Mutual Evaluation of Large Language Models
League of LLMs organizes LLMs into a self-governed mutual evaluation league using dynamic, transparent, objective, and professional criteria to distinguish model capabilities with 70.7% top-k ranking stability.
-
DataComp-LM: In search of the next generation of training sets for language models
DCLM-Baseline dataset lets a 7B model reach 64% 5-shot MMLU accuracy after 2.6T tokens, beating prior open-data models by 6.6 points on MMLU with 40% less compute.
-
Measuring Reasoning Quality in LLMs: A Multi-Dimensional Behavioral Framework
A multi-dimensional framework with six dimensions (Correctness, Consistency, Robustness, Logical Coherence, Efficiency, Stability) is applied to seven LLMs on 975 items, revealing orthogonality between logical coheren...
-
Measuring Reasoning Quality in LLMs: A Multi-Dimensional Behavioral Framework
A behavioral framework operationalizes six dimensions of LLM reasoning quality and shows they are largely independent from accuracy, revealing issues with single-metric evaluation.
-
Measuring Reasoning Quality in LLMs: A Multi-Dimensional Behavioral Framework
Proposes a multi-dimensional behavioral framework with six dimensions (Correctness, Consistency, Robustness, Local Logical Coherence, Efficiency, Stability) plus deployment-aware aggregation to diagnose LLM reasoning ...
-
The Illusion of Reasoning: Exposing Evasive Data Contamination in LLMs via Zero-CoT Truncation
ZCP detects direct and evasive data contamination in LLMs by truncating CoT reasoning and contrasting zero-CoT accuracy on original versus perturbed isomorphic datasets, plus a Contamination Confidence metric.
-
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.
-
LLM Benchmark Datasets Should Be Contamination-Resistant
Authors call for contamination-resistant LLM benchmarks that exploit Transformer training-inference asymmetry and require new mathematical methods for cross-architecture interoperability.
-
Measuring AI Reasoning: A Guide for Researchers
Reasoning in language models should be measured by the faithfulness and validity of their multi-step search processes and intermediate traces, not final-answer accuracy.
-
Compiled AI: Deterministic Code Generation for LLM-Based Workflow Automation
Compiled AI generates deterministic code artifacts from LLMs in a one-time compilation step, enabling reliable workflow execution with zero runtime tokens after break-even.
-
ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools
GLM-4 models rival or exceed GPT-4 on MMLU, GSM8K, MATH, BBH, GPQA, HumanEval, IFEval, long-context tasks, and Chinese alignment while adding autonomous tool use for web, code, and image generation.
-
Benchmark Data Contamination of Large Language Models: A Survey
A survey reviewing benchmark data contamination in LLMs, its impact on evaluation, and alternative assessment approaches.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.