Saved in:
Bibliographic Details
Main Authors: Ivanov, Igor, Volkov, Dmitrii
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.06738
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909486157922304
author Ivanov, Igor
Volkov, Dmitrii
author_facet Ivanov, Igor
Volkov, Dmitrii
contents Recent work showed that small changes in benchmark questions can reduce LLMs' reasoning and recall. We explore two such changes: pairing questions and adding more answer options, on three benchmarks: WMDP-bio, GPQA, and MMLU variants. We find that for more capable models, these predictably reduce performance, essentially heightening the performance ceiling of a benchmark and unsaturating it again. We suggest this approach can resurrect old benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2502_06738
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Resurrecting saturated LLM benchmarks with adversarial encoding
Ivanov, Igor
Volkov, Dmitrii
Machine Learning
Recent work showed that small changes in benchmark questions can reduce LLMs' reasoning and recall. We explore two such changes: pairing questions and adding more answer options, on three benchmarks: WMDP-bio, GPQA, and MMLU variants. We find that for more capable models, these predictably reduce performance, essentially heightening the performance ceiling of a benchmark and unsaturating it again. We suggest this approach can resurrect old benchmarks.
title Resurrecting saturated LLM benchmarks with adversarial encoding
topic Machine Learning
url https://arxiv.org/abs/2502.06738