Copertina

From Hallucination to Structure Snowballing: The Alignment Tax of Constrained Decoding in LLM Reflection

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autore principale:	Zhou, Hongxu
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Computation and Language
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Investigating and Mitigating the Multimodal Hallucination Snowballing in Large Vision-Language Models
di: Zhong, Weihong, et al.
Pubblicazione: (2024)

The Hallucination Tax of Reinforcement Finetuning
di: Song, Linxin, et al.
Pubblicazione: (2025)

Rethinking External Slow-Thinking: From Snowball Errors to Probability of Correct Reasoning
di: Gan, Zeyu, et al.
Pubblicazione: (2025)

The UNDO Flip-Flop: A Controlled Probe for Reversible Semantic State Management in State Space Model
di: Zhou, Hongxu
Pubblicazione: (2026)

Joint Effects of Argumentation Theory, Audio Modality and Data Enrichment on LLM-Based Fallacy Classification
di: Zhou, Hongxu, et al.
Pubblicazione: (2025)

MM-Snowball: Evaluating and Mitigating Hallucination Snowballing in Multimodal Multi-Turn Dialogue
di: Jiang, Yue, et al.
Pubblicazione: (2026)

Online Merging Optimizers for Boosting Rewards and Mitigating Tax in Alignment
di: Lu, Keming, et al.
Pubblicazione: (2024)

Transfer Q Star: Principled Decoding for LLM Alignment
di: Chakraborty, Souradip, et al.
Pubblicazione: (2024)

Flexible and Efficient Grammar-Constrained Decoding
di: Park, Kanghee, et al.
Pubblicazione: (2025)

Copy-as-Decode: Grammar-Constrained Parallel Prefill for LLM Editing
di: Liu, Ziyang
Pubblicazione: (2026)

Collab: Controlled Decoding using Mixture of Agents for LLM Alignment
di: Chakraborty, Souradip, et al.
Pubblicazione: (2025)

Reducing Tool Hallucination via Reliability Alignment
di: Xu, Hongshen, et al.
Pubblicazione: (2024)

Mitigating Hallucinations in Large Vision-Language Models with Internal Fact-based Contrastive Decoding
di: Wang, Chao, et al.
Pubblicazione: (2025)

MAD: Multi-Alignment MEG-to-Text Decoding
di: Yang, Yiqian, et al.
Pubblicazione: (2024)

Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection
di: Sun, Guanglong, et al.
Pubblicazione: (2026)

LLM Hallucination Detection: HSAD
di: Li, JinXin, et al.
Pubblicazione: (2025)

Shakespearean Sparks: The Dance of Hallucination and Creativity in LLMs' Decoding Layers
di: He, Zicong, et al.
Pubblicazione: (2025)

Draft-Conditioned Constrained Decoding for Structured Generation in LLMs
di: Reddy, Avinash, et al.
Pubblicazione: (2026)

Safety Alignment via Constrained Knowledge Unlearning
di: Shi, Zesheng, et al.
Pubblicazione: (2025)

TaxPraBen: A Scalable Benchmark for Structured Evaluation of LLMs in Chinese Real-World Tax Practice
di: Hu, Gang, et al.
Pubblicazione: (2026)

Alignment-Augmented Speculative Decoding with Alignment Sampling and Conditional Verification
di: Wang, Jikai, et al.
Pubblicazione: (2025)

Banishing LLM Hallucinations Requires Rethinking Generalization
di: Li, Johnny, et al.
Pubblicazione: (2024)

Constrained Decoding with Speculative Lookaheads
di: Nakshatri, Nishanth, et al.
Pubblicazione: (2024)

GRAD: Graph-Retrieved Adaptive Decoding for Hallucination Mitigation
di: Nguyen, Manh, et al.
Pubblicazione: (2025)

Grammar-Constrained Decoding for Structured NLP Tasks without Finetuning
di: Geng, Saibo, et al.
Pubblicazione: (2023)

Schema Key Wording as an Instruction Channel in Structured Generation under Constrained Decoding
di: Le, Yifan
Pubblicazione: (2026)

Reinforcement Learning with Semantic Rewards Enables Low-Resource Language Expansion without Alignment Tax
di: Su, Zeli, et al.
Pubblicazione: (2026)

MARCH: Multi-Agent Reinforced Self-Check for LLM Hallucination
di: Li, Zhuo, et al.
Pubblicazione: (2026)

Self-Alignment for Factuality: Mitigating Hallucinations in LLMs via Self-Evaluation
di: Zhang, Xiaoying, et al.
Pubblicazione: (2024)

Multi-Drafter Speculative Decoding with Alignment Feedback
di: Kim, Taehyeon, et al.
Pubblicazione: (2026)

The Format Tax
di: Lee, Ivan Yee, et al.
Pubblicazione: (2026)

Detecting Hallucinations in Authentic LLM-Human Interactions
di: Ren, Yujie, et al.
Pubblicazione: (2025)

HICD: Hallucination-Inducing via Attention Dispersion for Contrastive Decoding to Mitigate Hallucinations in Large Language Models
di: Jiang, Xinyan, et al.
Pubblicazione: (2025)

On-Policy Self-Alignment with Fine-grained Knowledge Feedback for Hallucination Mitigation
di: Wen, Xueru, et al.
Pubblicazione: (2024)

Disperse-Then-Merge: Pushing the Limits of Instruction Tuning via Alignment Tax Reduction
di: Fu, Tingchen, et al.
Pubblicazione: (2024)

Mitigating Multimodal Hallucinations via Gradient-based Self-Reflection
di: Wang, Shan, et al.
Pubblicazione: (2025)

GRIFFIN: Effective Token Alignment for Faster Speculative Decoding
di: Hu, Shijing, et al.
Pubblicazione: (2025)

FLAP: Flow-Adhering Planning with Constrained Decoding in LLMs
di: Roy, Shamik, et al.
Pubblicazione: (2024)

Decoding Alignment: A Critical Survey of LLM Development Initiatives through Value-setting and Data-centric Lens
di: Chalkidis, Ilias
Pubblicazione: (2025)

Tool-Call Dependency Structure is Linearly Decodable in LLM Agent Residual Streams
di: Sun, Tianda, et al.
Pubblicazione: (2026)