Salvato in:
Dettagli Bibliografici
Autore principale: Xia, Wei
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2510.23650
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915580111486976
author Xia, Wei
author_facet Xia, Wei
contents We proposed Static and Dynamic -- two zero-shot logits-layer debiasing methods. Dynamic reduces bias by up to 70% with minimal fluency loss. Logits intervention outperforms hidden-layer approaches. We show semantic-aware logits intervention is stable and effective for debiasing aligned LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2510_23650
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Beyond Hidden-Layer Manipulation: Semantically-Aware Logit Interventions for Debiasing LLMs
Xia, Wei
Machine Learning
Artificial Intelligence
We proposed Static and Dynamic -- two zero-shot logits-layer debiasing methods. Dynamic reduces bias by up to 70% with minimal fluency loss. Logits intervention outperforms hidden-layer approaches. We show semantic-aware logits intervention is stable and effective for debiasing aligned LLMs.
title Beyond Hidden-Layer Manipulation: Semantically-Aware Logit Interventions for Debiasing LLMs
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2510.23650