Image de couverture de livre

From Context Shift to Stylistic Collapse: Why Training Objectives Matter More Than Scale

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteur principal:	Mahapatra, Rohan
Format:	Preprint
Publié:	2026
Sujets:	Computation and Language
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Attention Reveals More Than Tokens: Training-Free Long-Context Reasoning with Attention-guided Retrieval
par: Zhang, Yuwei, et autres
Publié: (2025)

Long Is More Important Than Difficult for Training Reasoning Models
par: Shen, Si, et autres
Publié: (2025)

Learning Through Dialogue: Engagement and Efficacy Matter More Than Explanations
par: Furniturewala, Shaz, et autres
Publié: (2026)

The Algebra of Meaning: Why Machines Need Montague More Than Moore's Law
par: Jeong, Cheonkam, et autres
Publié: (2025)

Why Are Web AI Agents More Vulnerable Than Standalone LLMs? A Security Analysis
par: Chiang, Jeffrey Yang Fan, et autres
Publié: (2025)

Controllable Stylistic Text Generation with Train-Time Attribute-Regularized Diffusion
par: Zhou, Fan, et autres
Publié: (2025)

MoreHopQA: More Than Multi-hop Reasoning
par: Schnitzler, Julian, et autres
Publié: (2024)

More Than Sum of Its Parts: Deciphering Intent Shifts in Multimodal Hate Speech Detection
par: Sun, Runze, et autres
Publié: (2026)

More Rounds, More Noise: Why Multi-Turn Review Fails to Improve Cross-Context Verification
par: Tae-Eun, Song
Publié: (2026)

Why Low-Resource NLP Needs More Than Cross-Lingual Transfer: Lessons Learned from Luxembourgish
par: Philippy, Fred, et autres
Publié: (2026)

Where Matters More Than What: Decoding-aligned KV Cache Compression via Position-aware Pseudo Queries
par: Tian, Zhenxu, et autres
Publié: (2026)

Rethinking Layer Redundancy: Calibration Matters More Than Search in LLM Depth Pruning
par: Kim, Minkyu, et autres
Publié: (2026)

Narrative Flattening: How Post-Training Compresses Thematic, Affective, and Stylistic Variation in LLM Fiction
par: Li, Zehan, et autres
Publié: (2026)

Why LVLMs Are More Prone to Hallucinations in Longer Responses: The Role of Context
par: Zheng, Ge, et autres
Publié: (2025)

Confidence Should Be Calibrated More Than One Turn Deep
par: Zhang, Zhaohan, et autres
Publié: (2026)

Should LLM Safety Be More Than Refusing Harmful Instructions?
par: Maskey, Utsav, et autres
Publié: (2025)

Stylistic Evolution and LLM Neutrality in Singlish Language
par: Foo, Linus Tze En, et autres
Publié: (2026)

Detecting Stylistic Fingerprints of Large Language Models
par: Bitton, Yehonatan, et autres
Publié: (2025)

Neurobiber: Fast and Interpretable Stylistic Feature Extraction
par: Alkiek, Kenan, et autres
Publié: (2025)

Why Are Linear RNNs More Parallelizable?
par: Merrill, William, et autres
Publié: (2026)

Scaling with Collapse: Efficient and Predictable Training of LLM Families
par: Bergsma, Shane, et autres
Publié: (2025)

IDIOLEX: Unified and Continuous Representations for Idiolectal and Stylistic Variation
par: Kantharuban, Anjali, et autres
Publié: (2026)

When Large Language Models are More PersuasiveThan Incentivized Humans, and Why
par: Schoenegger, Philipp, et autres
Publié: (2025)

More Than Meets the Eye? Uncovering the Reasoning-Planning Disconnect in Training Vision-Language Driving Models
par: Song, Xurui, et autres
Publié: (2025)

Tokenization Is More Than Compression
par: Schmidt, Craig W., et autres
Publié: (2024)

More is not always better? Enhancing Many-Shot In-Context Learning with Differentiated and Reweighting Objectives
par: Zhang, Xiaoqing, et autres
Publié: (2025)

Are Any-to-Any Models More Consistent Across Modality Transfers Than Specialists?
par: Chung, Jiwan, et autres
Publié: (2025)

More Than Catastrophic Forgetting: Integrating General Capabilities For Domain-Specific LLMs
par: Liu, Chengyuan, et autres
Publié: (2024)

Why Slop Matters
par: Kommers, Cody, et autres
Publié: (2025)

LLMs Struggle with Abstract Meaning Comprehension More Than Expected
par: Alhazmi, Hamoud, et autres
Publié: (2026)

CLASE: A Hybrid Method for Chinese Legalese Stylistic Evaluation
par: Ma, Yiran Rex, et autres
Publié: (2026)

Latent Space Interpretation for Stylistic Analysis and Explainable Authorship Attribution
par: Alshomary, Milad, et autres
Publié: (2024)

ShortGPT: Layers in Large Language Models are More Redundant Than You Expect
par: Men, Xin, et autres
Publié: (2024)

More Than a Score: Probing the Impact of Prompt Specificity on LLM Code Generation
par: Zi, Yangtian, et autres
Publié: (2025)

Why Mean Pooling Works: Quantifying Second-Order Collapse in Text Embeddings
par: Hara, Tomomasa, et autres
Publié: (2026)

Annotations Mitigate Post-Training Mode Collapse
par: Springer, Jacob Mitchell, et autres
Publié: (2026)

Training-Free Long-Context Scaling of Large Language Models
par: An, Chenxin, et autres
Publié: (2024)

Real World Conversational Entity Linking Requires More Than Zeroshots
par: Hoveyda, Mohanna, et autres
Publié: (2024)

Towards a Linguistic Evaluation of Narratives: A Quantitative Stylistic Framework
par: Maisto, Alessandro
Publié: (2026)

StylusAI: Stylistic Adaptation for Robust German Handwritten Text Generation
par: Riaz, Nauman, et autres
Publié: (2024)