@inproceedings{wang-etal-2026-benchmarks,
title = "Do Benchmarks Measure the Same Thing? Evaluation-Free Benchmark Redundancy Analysis with {SAE}",
author = "Wang, Yaoning and
Ying, Jiahao and
Ma, Yubo and
Wan, Yu and
Liu, Dayiheng and
Cao, Yixin and
Jiang, Yu-Gang",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1284/",
pages = "29255--29271",
ISBN = "979-8-89176-459-0",
abstract = "The rapid growth of LLM evaluation resources makes exhaustive benchmark curation increasingly difficult. Existing efficient benchmarking methods reduce evaluation cost by selecting representative subsets or predicting full-benchmark performance, but typically require evaluation results from many models and additional training. In this paper, we ask a more fundamental question: to what extent do benchmarks actually measure different things? We study this question at two levels: intra-benchmark redundancy, which supports benchmark compression, and inter-benchmark similarity, which supports evaluation-suite design. We first formalize both notions through a functionalist behavioral criterion, and then propose an evaluation-free proxy based on Sparse Autoencoder (SAE) features for redundancy diagnosis. Our central hypothesis is that behavioral overlap can be approximated by feature overlap. This hypothesis yields level-specific metrics based on feature-coverage saturation within benchmarks and feature-footprint overlap across benchmarks. Experiments show that SAE feature-based redundancy strongly correlates with performance-based redundancy across 17 benchmarks, while SAE-guided subset selection produces more stable and representative subsets than random sampling. SAE feature overlap also predicts inter-benchmark performance similarity across 8 benchmarks. These results show that SAE features can turn benchmark analysis from a costly performance-matrix problem into a feature-footprint problem, enabling evaluation-free benchmark compression and suite curation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-benchmarks">
<titleInfo>
<title>Do Benchmarks Measure the Same Thing? Evaluation-Free Benchmark Redundancy Analysis with SAE</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yaoning</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiahao</namePart>
<namePart type="family">Ying</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yubo</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Wan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dayiheng</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yixin</namePart>
<namePart type="family">Cao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yu-Gang</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>The rapid growth of LLM evaluation resources makes exhaustive benchmark curation increasingly difficult. Existing efficient benchmarking methods reduce evaluation cost by selecting representative subsets or predicting full-benchmark performance, but typically require evaluation results from many models and additional training. In this paper, we ask a more fundamental question: to what extent do benchmarks actually measure different things? We study this question at two levels: intra-benchmark redundancy, which supports benchmark compression, and inter-benchmark similarity, which supports evaluation-suite design. We first formalize both notions through a functionalist behavioral criterion, and then propose an evaluation-free proxy based on Sparse Autoencoder (SAE) features for redundancy diagnosis. Our central hypothesis is that behavioral overlap can be approximated by feature overlap. This hypothesis yields level-specific metrics based on feature-coverage saturation within benchmarks and feature-footprint overlap across benchmarks. Experiments show that SAE feature-based redundancy strongly correlates with performance-based redundancy across 17 benchmarks, while SAE-guided subset selection produces more stable and representative subsets than random sampling. SAE feature overlap also predicts inter-benchmark performance similarity across 8 benchmarks. These results show that SAE features can turn benchmark analysis from a costly performance-matrix problem into a feature-footprint problem, enabling evaluation-free benchmark compression and suite curation.</abstract>
<identifier type="citekey">wang-etal-2026-benchmarks</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1284/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>29255</start>
<end>29271</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Do Benchmarks Measure the Same Thing? Evaluation-Free Benchmark Redundancy Analysis with SAE
%A Wang, Yaoning
%A Ying, Jiahao
%A Ma, Yubo
%A Wan, Yu
%A Liu, Dayiheng
%A Cao, Yixin
%A Jiang, Yu-Gang
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F wang-etal-2026-benchmarks
%X The rapid growth of LLM evaluation resources makes exhaustive benchmark curation increasingly difficult. Existing efficient benchmarking methods reduce evaluation cost by selecting representative subsets or predicting full-benchmark performance, but typically require evaluation results from many models and additional training. In this paper, we ask a more fundamental question: to what extent do benchmarks actually measure different things? We study this question at two levels: intra-benchmark redundancy, which supports benchmark compression, and inter-benchmark similarity, which supports evaluation-suite design. We first formalize both notions through a functionalist behavioral criterion, and then propose an evaluation-free proxy based on Sparse Autoencoder (SAE) features for redundancy diagnosis. Our central hypothesis is that behavioral overlap can be approximated by feature overlap. This hypothesis yields level-specific metrics based on feature-coverage saturation within benchmarks and feature-footprint overlap across benchmarks. Experiments show that SAE feature-based redundancy strongly correlates with performance-based redundancy across 17 benchmarks, while SAE-guided subset selection produces more stable and representative subsets than random sampling. SAE feature overlap also predicts inter-benchmark performance similarity across 8 benchmarks. These results show that SAE features can turn benchmark analysis from a costly performance-matrix problem into a feature-footprint problem, enabling evaluation-free benchmark compression and suite curation.
%U https://aclanthology.org/2026.emnlp-main.1284/
%P 29255-29271
Markdown (Informal)
[Do Benchmarks Measure the Same Thing? Evaluation-Free Benchmark Redundancy Analysis with SAE](https://aclanthology.org/2026.emnlp-main.1284/) (Wang et al., EMNLP 2026)
ACL