Imagen de Portada

Phonetic Perturbations Reveal Tokenizer-Rooted Safety Gaps in LLMs

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Aswal, Darpan, Jaiswal, Siddharth D
Formato:	Preprint
Publicado:	2025
Materias:	Computation and Language Artificial Intelligence
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

ConceptGuard: Neuro-Symbolic Safety Guardrails via Sparse Interpretable Jailbreak Concepts
por: Aswal, Darpan, et al.
Publicado: (2025)

Efficient Environmental Claim Detection with Hyperbolic Graph Neural Networks
por: Aswal, Darpan, et al.
Publicado: (2025)

Morphemes Without Borders: Evaluating Root-Pattern Morphology in Arabic Tokenizers and LLMs
por: Alakeel, Yara, et al.
Publicado: (2026)

LLMs Know More Than Words: A Genre Study with Syntax, Metaphor & Phonetics
por: Shi, Weiye, et al.
Publicado: (2025)

Training LLMs Beyond Next Token Prediction -- Filling the Mutual Information Gap
por: Yang, Chun-Hao, et al.
Publicado: (2025)

LLMs know their vulnerabilities: Uncover Safety Gaps through Natural Distribution Shifts
por: Ren, Qibing, et al.
Publicado: (2024)

Efficient Switchable Safety Control in LLMs via Magic-Token-Guided Co-Training
por: Si, Jianfeng, et al.
Publicado: (2025)

Are LLMs good pragmatic speakers?
por: Jian, Mingyue, et al.
Publicado: (2024)

Multitask Mayhem: Unveiling and Mitigating Safety Gaps in LLMs Fine-tuning
por: Jan, Essa, et al.
Publicado: (2024)

Phonetically rich corpus construction for a low-resourced language
por: Amadeus, Marcellus, et al.
Publicado: (2024)

LLMs are Not Just Next Token Predictors
por: Downes, Stephen M., et al.
Publicado: (2024)

Hessian-Enhanced Token Attribution (HETA): Interpreting Autoregressive LLMs
por: Pramanik, Vishal, et al.
Publicado: (2026)

Revealing the Numeracy Gap: An Empirical Investigation of Text Embedding Models
por: Deng, Ningyuan, et al.
Publicado: (2025)

Echoes of Phonetics: Unveiling Relevant Acoustic Cues for ASR via Feature Attribution
por: Fucci, Dennis, et al.
Publicado: (2025)

TECP: Token-Entropy Conformal Prediction for LLMs
por: Xu, Beining, et al.
Publicado: (2025)

From Tokens to Words: On the Inner Lexicon of LLMs
por: Kaplan, Guy, et al.
Publicado: (2024)

Token Statistics Reveal Conversational Drift in Multi-turn LLM Interaction
por: Hafez, Wael, et al.
Publicado: (2026)

Autonomous Evaluation of LLMs for Truth Maintenance and Reasoning Tasks
por: Karia, Rushang, et al.
Publicado: (2024)

HealthQA-BR: A System-Wide Benchmark Reveals Critical Knowledge Gaps in Large Language Models
por: D'addario, Andrew Maranhão Ventura
Publicado: (2025)

PERCORE: A Deep Learning-Based Framework for Persian Spelling Correction with Phonetic Analysis
por: Dashti, Seyed Mohammad Sadegh, et al.
Publicado: (2024)

Attributional Safety Failures in Large Language Models under Code-Mixed Perturbations
por: Banerjee, Somnath, et al.
Publicado: (2025)

Bridging the Creativity Understanding Gap: Small-Scale Human Alignment Enables Expert-Level Humor Ranking in LLMs
por: Zhou, Kuan Lok, et al.
Publicado: (2025)

TokenSkip: Controllable Chain-of-Thought Compression in LLMs
por: Xia, Heming, et al.
Publicado: (2025)

Discrete Tokenization for Multimodal LLMs: A Comprehensive Survey
por: Li, Jindong, et al.
Publicado: (2025)

Say Anything but This: When Tokenizer Betrays Reasoning in LLMs
por: Ayoobi, Navid, et al.
Publicado: (2026)

Improving Speech Recognition of Named Entities in Classroom Speech with LLM Revision and Phonetic-Semantic Context
por: Trinh, Viet Anh, et al.
Publicado: (2025)

Speak & Spell: LLM-Driven Controllable Phonetic Error Augmentation for Robust Dialogue State Tracking
por: Lee, Jihyun, et al.
Publicado: (2024)

Causal Interventions Reveal Shared Structure Across English Filler-Gap Constructions
por: Boguraev, Sasha, et al.
Publicado: (2025)

The Instruction Gap: LLMs get lost in Following Instruction
por: Tripathi, Vishesh, et al.
Publicado: (2025)

Mind the (Belief) Gap: Group Identity in the World of LLMs
por: Borah, Angana, et al.
Publicado: (2025)

An Expert is Worth One Token: Synergizing Multiple Expert LLMs as Generalist via Expert Token Routing
por: Chai, Ziwei, et al.
Publicado: (2024)

Normalization through Fine-tuning: Understanding Wav2vec 2.0 Embeddings for Phonetic Analysis
por: Wang, Yiming, et al.
Publicado: (2025)

TIDE: Every Layer Knows the Token Beneath the Context
por: Jaiswal, Ajay, et al.
Publicado: (2026)

SelecTKD: Selective Token-Weighted Knowledge Distillation for LLMs
por: Huang, Haiduo, et al.
Publicado: (2025)

Phonetically-Augmented Discriminative Rescoring for Voice Search Error Correction
por: Van Gysel, Christophe, et al.
Publicado: (2025)

MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbations
por: Huang, Kaixuan, et al.
Publicado: (2025)

Tokenization Constraints in LLMs: A Study of Symbolic and Arithmetic Reasoning Limits
por: Zhang, Xiang, et al.
Publicado: (2025)

Adaptive Token Boundaries: Integrating Human Chunking Mechanisms into Multimodal LLMs
por: Yu, Dongxing
Publicado: (2025)

Informed Routing in LLMs: Smarter Token-Level Computation for Faster Inference
por: Han, Chao, et al.
Publicado: (2025)

STAPO: Stabilizing Reinforcement Learning for LLMs by Silencing Rare Spurious Tokens
por: Liu, Shiqi, et al.
Publicado: (2026)