Salvato in:
Dettagli Bibliografici
Autori principali: Panda, Sailesh, Kadasi, Pritam, Upperwal, Abhishek, Singh, Mayank
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:https://arxiv.org/abs/2605.00817
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916042104635392
author Panda, Sailesh
Kadasi, Pritam
Upperwal, Abhishek
Singh, Mayank
author_facet Panda, Sailesh
Kadasi, Pritam
Upperwal, Abhishek
Singh, Mayank
contents Large language models (LLMs) often achieve strong performance on reasoning benchmarks, but final-answer accuracy alone does not show whether they faithfully execute the procedure specified in a prompt. We introduce a controlled diagnostic benchmark for procedural execution, where models are given a step-wise arithmetic procedure and two numeric inputs, and must return the final computed value. Complexity is varied through procedure length and look-back dependencies over intermediate variables. Average first-answer accuracy drops from 63% on 5-step procedures to 20% on 95-step procedures. Generation-level analysis shows that failures often involve missing answers, premature answers, self-correction after an initial error and under-executed traces. These findings suggest that apparent reasoning ability can mask substantial weaknesses in faithful long-horizon procedural execution.
format Preprint
id arxiv_https___arxiv_org_abs_2605_00817
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle When LLMs Stop Following Steps: A Diagnostic Study of Procedural Execution in Language Models
Panda, Sailesh
Kadasi, Pritam
Upperwal, Abhishek
Singh, Mayank
Computation and Language
Large language models (LLMs) often achieve strong performance on reasoning benchmarks, but final-answer accuracy alone does not show whether they faithfully execute the procedure specified in a prompt. We introduce a controlled diagnostic benchmark for procedural execution, where models are given a step-wise arithmetic procedure and two numeric inputs, and must return the final computed value. Complexity is varied through procedure length and look-back dependencies over intermediate variables. Average first-answer accuracy drops from 63% on 5-step procedures to 20% on 95-step procedures. Generation-level analysis shows that failures often involve missing answers, premature answers, self-correction after an initial error and under-executed traces. These findings suggest that apparent reasoning ability can mask substantial weaknesses in faithful long-horizon procedural execution.
title When LLMs Stop Following Steps: A Diagnostic Study of Procedural Execution in Language Models
topic Computation and Language
url https://arxiv.org/abs/2605.00817