Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Hoshino, Sho, Honda, Ukyo, Zhang, Peinan
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2604.19395
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913051080392704
author Hoshino, Sho
Honda, Ukyo
Zhang, Peinan
author_facet Hoshino, Sho
Honda, Ukyo
Zhang, Peinan
contents While self-consistency is known to improve performance on symbolic reasoning, its effect on the recall of encyclopedic knowledge is unclear due to a lack of targeted evaluation grounds. To address this, we establish such a knowledge recall split for the popular MMLU benchmark by applying a data-driven heuristic from prior work. We validate this split by showing that the performance patterns on the symbolic reasoning and knowledge recall subsets mirror those of GSM8K and MedMCQA, respectively. Using this solid ground, we find that self-consistency consistently improves performance across both symbolic reasoning and knowledge recall, even though its underlying CoT prompting is primarily effective for symbolic reasoning. As a result, we achieve an 89\% accuracy on MMLU, the best performance to date with the use of GPT-4o.
format Preprint
id arxiv_https___arxiv_org_abs_2604_19395
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Does Self-Consistency Improve the Recall of Encyclopedic Knowledge?
Hoshino, Sho
Honda, Ukyo
Zhang, Peinan
Computation and Language
While self-consistency is known to improve performance on symbolic reasoning, its effect on the recall of encyclopedic knowledge is unclear due to a lack of targeted evaluation grounds. To address this, we establish such a knowledge recall split for the popular MMLU benchmark by applying a data-driven heuristic from prior work. We validate this split by showing that the performance patterns on the symbolic reasoning and knowledge recall subsets mirror those of GSM8K and MedMCQA, respectively. Using this solid ground, we find that self-consistency consistently improves performance across both symbolic reasoning and knowledge recall, even though its underlying CoT prompting is primarily effective for symbolic reasoning. As a result, we achieve an 89\% accuracy on MMLU, the best performance to date with the use of GPT-4o.
title Does Self-Consistency Improve the Recall of Encyclopedic Knowledge?
topic Computation and Language
url https://arxiv.org/abs/2604.19395