Saved in:
Bibliographic Details
Main Authors: Stringli, Elena, Lymperaiou, Maria, Filandrianos, Giorgos, Voulodimos, Athanasios, Stamou, Giorgos
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.12821
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912407752802304
author Stringli, Elena
Lymperaiou, Maria
Filandrianos, Giorgos
Voulodimos, Athanasios
Stamou, Giorgos
author_facet Stringli, Elena
Lymperaiou, Maria
Filandrianos, Giorgos
Voulodimos, Athanasios
Stamou, Giorgos
contents Inverse tasks can uncover potential reasoning gaps as Large Language Models (LLMs) scale up. In this work, we explore the redefinition task, in which we assign alternative values to well-known physical constants and units of measure, prompting LLMs to respond accordingly. Our findings show that not only does model performance degrade with scale, but its false confidence also rises. Moreover, while factors such as prompting strategies or response formatting are influential, they do not preclude LLMs from anchoring to memorized values.
format Preprint
id arxiv_https___arxiv_org_abs_2502_12821
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Pitfalls of Scale: Investigating the Inverse Task of Redefinition in Large Language Models
Stringli, Elena
Lymperaiou, Maria
Filandrianos, Giorgos
Voulodimos, Athanasios
Stamou, Giorgos
Computation and Language
Inverse tasks can uncover potential reasoning gaps as Large Language Models (LLMs) scale up. In this work, we explore the redefinition task, in which we assign alternative values to well-known physical constants and units of measure, prompting LLMs to respond accordingly. Our findings show that not only does model performance degrade with scale, but its false confidence also rises. Moreover, while factors such as prompting strategies or response formatting are influential, they do not preclude LLMs from anchoring to memorized values.
title Pitfalls of Scale: Investigating the Inverse Task of Redefinition in Large Language Models
topic Computation and Language
url https://arxiv.org/abs/2502.12821