Bridging the Know-Act Gap via Task-Level Autoregressive Reasoning

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ahn, Jihyun Janice, Kamoi, Ryo, Atil, Berk, Lou, Renze, Kang, WonWoo, Park, Heehyun, Das, Sarkar Snigdha Sarathi, Zou, Zhuoyang, Lu, Xiaoxin, Zhang, Yusen, Shah, Asfahan, Tanvir, Ridwanul Hasan, Zhao, Lingxiao, Huang, Hongxi, Venkatesh, Vignesh, Lin, Dianjun, Shah, Hamid, Wang, Wentao, Song, Zhanpeng, Bassin, Joshua Reed, Patel, Dax, Agrahar, Ishan Appareddy, Pardasani, Sahil, Dong, Xin, Rahbari, Fatemeh, Rishel, Benjamin David, Lee, Soochan Andrew, Boghani, Yuv, AlNaseeb, Ali B., Suby, Pranav, Bae, Seokhyeon, Buddharaju, Shreya, Kula, Damien, Das, Soumyadeep, Liu, Hanyang Frank, Mo, Faye, Yin, Wenpeng
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917359261843456
author Ahn, Jihyun Janice
Kamoi, Ryo
Atil, Berk
Lou, Renze
Kang, WonWoo
Park, Heehyun
Das, Sarkar Snigdha Sarathi
Zou, Zhuoyang
Lu, Xiaoxin
Zhang, Yusen
Shah, Asfahan
Tanvir, Ridwanul Hasan
Zhao, Lingxiao
Huang, Hongxi
Venkatesh, Vignesh
Lin, Dianjun
Shah, Hamid
Wang, Wentao
Song, Zhanpeng
Bassin, Joshua Reed
Patel, Dax
Agrahar, Ishan Appareddy
Pardasani, Sahil
Dong, Xin
Rahbari, Fatemeh
Rishel, Benjamin David
Lee, Soochan Andrew
Boghani, Yuv
AlNaseeb, Ali B.
Suby, Pranav
Bae, Seokhyeon
Buddharaju, Shreya
Kula, Damien
Das, Soumyadeep
Liu, Hanyang Frank
Mo, Faye
Yin, Wenpeng
author_facet Ahn, Jihyun Janice
Kamoi, Ryo
Atil, Berk
Lou, Renze
Kang, WonWoo
Park, Heehyun
Das, Sarkar Snigdha Sarathi
Zou, Zhuoyang
Lu, Xiaoxin
Zhang, Yusen
Shah, Asfahan
Tanvir, Ridwanul Hasan
Zhao, Lingxiao
Huang, Hongxi
Venkatesh, Vignesh
Lin, Dianjun
Shah, Hamid
Wang, Wentao
Song, Zhanpeng
Bassin, Joshua Reed
Patel, Dax
Agrahar, Ishan Appareddy
Pardasani, Sahil
Dong, Xin
Rahbari, Fatemeh
Rishel, Benjamin David
Lee, Soochan Andrew
Boghani, Yuv
AlNaseeb, Ali B.
Suby, Pranav
Bae, Seokhyeon
Buddharaju, Shreya
Kula, Damien
Das, Soumyadeep
Liu, Hanyang Frank
Mo, Faye
Yin, Wenpeng
contents LLMs often generate seemingly valid answers to flawed or ill-posed inputs. This is not due to missing knowledge: under discriminative prompting, the same models can mostly identify such issues, yet fail to reflect this in standard generative responses. This reveals a fundamental know-act gap between discriminative recognition and generative behavior. Prior work largely characterizes this issue in narrow settings, such as math word problems or question answering, with limited focus on how to integrate these two modes. In this work, we present a comprehensive analysis using FaultyScience, a newly constructed large-scale, cross-disciplinary benchmark of faulty scientific questions. We show that the gap is pervasive and stems from token-level autoregression, which entangles task selection (validate vs. answer) with content generation, preventing discriminative knowledge from being utilized. To address this, we propose DeIllusionLLM, a task-level autoregressive framework that explicitly models this decision. Through self-distillation, the model unifies discriminative judgment and generative reasoning within a single backbone. Empirically, DeIllusionLLM substantially reduces answer-despite-error failures under natural prompting while maintaining general reasoning performance, demonstrating that self-distillation is an effective and scalable solution for bridging the discriminative-generative know-act gap
format Preprint
id arxiv_https___arxiv_org_abs_2603_22619
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Bridging the Know-Act Gap via Task-Level Autoregressive Reasoning
Ahn, Jihyun Janice
Kamoi, Ryo
Atil, Berk
Lou, Renze
Kang, WonWoo
Park, Heehyun
Das, Sarkar Snigdha Sarathi
Zou, Zhuoyang
Lu, Xiaoxin
Zhang, Yusen
Shah, Asfahan
Tanvir, Ridwanul Hasan
Zhao, Lingxiao
Huang, Hongxi
Venkatesh, Vignesh
Lin, Dianjun
Shah, Hamid
Wang, Wentao
Song, Zhanpeng
Bassin, Joshua Reed
Patel, Dax
Agrahar, Ishan Appareddy
Pardasani, Sahil
Dong, Xin
Rahbari, Fatemeh
Rishel, Benjamin David
Lee, Soochan Andrew
Boghani, Yuv
AlNaseeb, Ali B.
Suby, Pranav
Bae, Seokhyeon
Buddharaju, Shreya
Kula, Damien
Das, Soumyadeep
Liu, Hanyang Frank
Mo, Faye
Yin, Wenpeng
Artificial Intelligence
LLMs often generate seemingly valid answers to flawed or ill-posed inputs. This is not due to missing knowledge: under discriminative prompting, the same models can mostly identify such issues, yet fail to reflect this in standard generative responses. This reveals a fundamental know-act gap between discriminative recognition and generative behavior. Prior work largely characterizes this issue in narrow settings, such as math word problems or question answering, with limited focus on how to integrate these two modes. In this work, we present a comprehensive analysis using FaultyScience, a newly constructed large-scale, cross-disciplinary benchmark of faulty scientific questions. We show that the gap is pervasive and stems from token-level autoregression, which entangles task selection (validate vs. answer) with content generation, preventing discriminative knowledge from being utilized. To address this, we propose DeIllusionLLM, a task-level autoregressive framework that explicitly models this decision. Through self-distillation, the model unifies discriminative judgment and generative reasoning within a single backbone. Empirically, DeIllusionLLM substantially reduces answer-despite-error failures under natural prompting while maintaining general reasoning performance, demonstrating that self-distillation is an effective and scalable solution for bridging the discriminative-generative know-act gap
title Bridging the Know-Act Gap via Task-Level Autoregressive Reasoning
topic Artificial Intelligence
url https://arxiv.org/abs/2603.22619