First Heuristic Then Rational: Dynamic Use of Heuristics in Language Model Reasoning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Aoki, Yoichi, Kudo, Keito, Kuribayashi, Tatsuki, Sone, Shusaku, Taniguchi, Masaya, Sakaguchi, Keisuke, Inui, Kentaro
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916425054027776
author Aoki, Yoichi
Kudo, Keito
Kuribayashi, Tatsuki
Sone, Shusaku
Taniguchi, Masaya
Sakaguchi, Keisuke
Inui, Kentaro
author_facet Aoki, Yoichi
Kudo, Keito
Kuribayashi, Tatsuki
Sone, Shusaku
Taniguchi, Masaya
Sakaguchi, Keisuke
Inui, Kentaro
contents Multi-step reasoning instruction, such as chain-of-thought prompting, is widely adopted to explore better language models (LMs) performance. We report on the systematic strategy that LMs employ in such a multi-step reasoning process. Our controlled experiments reveal that LMs rely more heavily on heuristics, such as lexical overlap, in the earlier stages of reasoning, where more reasoning steps remain to reach a goal. Conversely, their reliance on heuristics decreases as LMs progress closer to the final answer through multiple reasoning steps. This suggests that LMs can backtrack only a limited number of future steps and dynamically combine heuristic strategies with rationale ones in tasks involving multi-step reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2406_16078
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle First Heuristic Then Rational: Dynamic Use of Heuristics in Language Model Reasoning
Aoki, Yoichi
Kudo, Keito
Kuribayashi, Tatsuki
Sone, Shusaku
Taniguchi, Masaya
Sakaguchi, Keisuke
Inui, Kentaro
Computation and Language
Multi-step reasoning instruction, such as chain-of-thought prompting, is widely adopted to explore better language models (LMs) performance. We report on the systematic strategy that LMs employ in such a multi-step reasoning process. Our controlled experiments reveal that LMs rely more heavily on heuristics, such as lexical overlap, in the earlier stages of reasoning, where more reasoning steps remain to reach a goal. Conversely, their reliance on heuristics decreases as LMs progress closer to the final answer through multiple reasoning steps. This suggests that LMs can backtrack only a limited number of future steps and dynamically combine heuristic strategies with rationale ones in tasks involving multi-step reasoning.
title First Heuristic Then Rational: Dynamic Use of Heuristics in Language Model Reasoning
topic Computation and Language
url https://arxiv.org/abs/2406.16078