Autoregressive Anchoring: Quantifying Numeric Persistence in Large Language Model Generation

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Authors: Архитект, VA
Format: Recurso digital
Published: Zenodo 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866901830963822592
author Архитект
VA
author_facet Архитект
VA
contents <blockquote> <p>This study quantifies the "echo" of numeric primes in Large Language Models. We demonstrate that anchoring is not merely a localized error but a decaying signal that propagates through the autoregressive generation process in LLM.</p> </blockquote>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18728819
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Autoregressive Anchoring: Quantifying Numeric Persistence in Large Language Model Generation
Архитект
VA
Large Language Machines
<blockquote> <p>This study quantifies the "echo" of numeric primes in Large Language Models. We demonstrate that anchoring is not merely a localized error but a decaying signal that propagates through the autoregressive generation process in LLM.</p> </blockquote>
title Autoregressive Anchoring: Quantifying Numeric Persistence in Large Language Model Generation
topic Large Language Machines
url https://doi.org/10.5281/zenodo.18728819