Varying Shades of Wrong: Aligning LLMs with Wrong Answers Only

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Yao, Jihan, Ding, Wenxuan, Feng, Shangbin, Wang, Lucy Lu, Tsvetkov, Yulia
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913547026432000
author Yao, Jihan
Ding, Wenxuan
Feng, Shangbin
Wang, Lucy Lu
Tsvetkov, Yulia
author_facet Yao, Jihan
Ding, Wenxuan
Feng, Shangbin
Wang, Lucy Lu
Tsvetkov, Yulia
contents In the absence of abundant reliable annotations for challenging tasks and contexts, how can we expand the frontier of LLM capabilities with potentially wrong answers? We focus on two research questions: (1) Can LLMs generate reliable preferences among wrong options? And if so, (2) Would alignment with such wrong-over-wrong preferences be helpful? We employ methods based on self-consistency, token probabilities, and LLM-as-a-judge to elicit wrong-over-wrong preferences, and fine-tune language models with preference optimization approaches using these synthesized preferences. Extensive experiments with seven LLMs and eight datasets demonstrate that (1) LLMs do have preliminary capability in distinguishing various shades of wrong, achieving up to 20.9% higher performance than random guess; (2) Alignment with wrong-over-wrong preferences helps LLMs to produce less wrong and sometimes even outright correct answers, while overall improving model calibration.
format Preprint
id arxiv_https___arxiv_org_abs_2410_11055
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Varying Shades of Wrong: Aligning LLMs with Wrong Answers Only
Yao, Jihan
Ding, Wenxuan
Feng, Shangbin
Wang, Lucy Lu
Tsvetkov, Yulia
Computation and Language
Artificial Intelligence
In the absence of abundant reliable annotations for challenging tasks and contexts, how can we expand the frontier of LLM capabilities with potentially wrong answers? We focus on two research questions: (1) Can LLMs generate reliable preferences among wrong options? And if so, (2) Would alignment with such wrong-over-wrong preferences be helpful? We employ methods based on self-consistency, token probabilities, and LLM-as-a-judge to elicit wrong-over-wrong preferences, and fine-tune language models with preference optimization approaches using these synthesized preferences. Extensive experiments with seven LLMs and eight datasets demonstrate that (1) LLMs do have preliminary capability in distinguishing various shades of wrong, achieving up to 20.9% higher performance than random guess; (2) Alignment with wrong-over-wrong preferences helps LLMs to produce less wrong and sometimes even outright correct answers, while overall improving model calibration.
title Varying Shades of Wrong: Aligning LLMs with Wrong Answers Only
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2410.11055