Copertina

How Does Unfaithful Reasoning Emerge from Autoregressive Training? A Study of Synthetic Experiments

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Wang, Fuxin, Alazali, Amr, Zhong, Yiqiao
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Machine Learning Artificial Intelligence I.2.6; I.2.7
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Shattered Compositionality: Counterintuitive Learning Dynamics of Transformers for Arithmetic
di: Zhao, Xingyu, et al.
Pubblicazione: (2026)

The Geometry of Thought: How Scale Restructures Reasoning In Large Language Models
di: Anderson, Samuel Cyrenius
Pubblicazione: (2026)

Entropy-Based Measurement of Value Drift and Alignment Work in Large Language Models
di: Fadli, Samih
Pubblicazione: (2025)

Overclocking LLM Reasoning: Monitoring and Controlling Thinking Path Lengths in LLMs
di: Eisenstadt, Roy, et al.
Pubblicazione: (2025)

CircuitProbe: Predicting Reasoning Circuits in Transformers via Stability Zone Detection
di: Panuganti, Rajkiran
Pubblicazione: (2026)

On Semantic Loss Fine-Tuning Approach for Preventing Model Collapse in Causal Reasoning
di: Deshmukh, Pratik, et al.
Pubblicazione: (2026)

Reasoning Large Language Model Errors Arise from Hallucinating Critical Problem Features
di: Heyman, Alex, et al.
Pubblicazione: (2025)

SCULPT: Constraint-Guided Pruned MCTS that Carves Efficient Paths for Mathematical Reasoning
di: Fang, Qitong, et al.
Pubblicazione: (2026)

Scaling Trends for Multi-Hop Contextual Reasoning in Mid-Scale Language Models
di: Steele, Brady, et al.
Pubblicazione: (2026)

Social Cooperation in Conversational AI Agents
di: Çelikok, Mustafa Mert, et al.
Pubblicazione: (2025)

FastForward Pruning: Efficient LLM Pruning via Single-Step Reinforcement Learning
di: Yuan, Xin, et al.
Pubblicazione: (2025)

Extreme AutoML: Analysis of Classification, Regression, and NLP Performance
di: Ratner, Edward, et al.
Pubblicazione: (2024)

PRPO: Aligning Process Reward with Outcome Reward in Policy Optimization
di: Ding, Ruiyi, et al.
Pubblicazione: (2026)

Quantization Undoes Alignment: Bias Emergence in Compressed LLMs Across Models and Precision Levels
di: Rath, Plawan Kumar, et al.
Pubblicazione: (2026)

OFMU: Optimization-Driven Framework for Machine Unlearning
di: Asif, Sadia, et al.
Pubblicazione: (2025)

BLOCK-EM: Preventing Emergent Misalignment via Latent Blocking
di: Ustaomeroglu, Muhammed, et al.
Pubblicazione: (2026)

Persona Features Control Emergent Misalignment
di: Wang, Miles, et al.
Pubblicazione: (2025)

Autonomous Deep Agent
di: Yu, Amy, et al.
Pubblicazione: (2025)

The Anti-Ouroboros Effect: Emergent Resilience in Large Language Models from Recursive Selective Feedback
di: Adapala, Sai Teja Reddy
Pubblicazione: (2025)

FastGRPO: Accelerating Policy Optimization via Concurrency-aware Speculative Decoding and Online Draft Learning
di: Zhang, Yizhou, et al.
Pubblicazione: (2025)

Task-Conditioned Routing Signatures in Sparse Mixture-of-Experts Transformers
di: Avinash, Mynampati Sri Ranganadha
Pubblicazione: (2026)

Node-Level Uncertainty Estimation in LLM-Generated SQL
di: Hasson, Hilaf, et al.
Pubblicazione: (2025)

ReflexGrad: Within-Episode Failure Recovery in LLM Agents via Progress-Gated Dual-Process Routing
di: Kadu, Ankush, et al.
Pubblicazione: (2025)

Benefits and Limitations of Communication in Multi-Agent Reasoning
di: Rizvi-Martel, Michael, et al.
Pubblicazione: (2025)

Planning vs Reasoning: Ablations to Test Capabilities of LoRA layers
di: Redkar, Neel
Pubblicazione: (2024)

Contextual Integrity in LLMs via Reasoning and Reinforcement Learning
di: Lan, Guangchen, et al.
Pubblicazione: (2025)

BitCal-TTS: Bit-Calibrated Test-Time Scaling for Quantized Reasoning Models
di: Patarlapalli, Sai Babu, et al.
Pubblicazione: (2026)

Beyond Pass@k: Breadth-Depth Metrics for Reasoning Boundaries
di: Dragoi, Marius, et al.
Pubblicazione: (2025)

Counterfactual Likelihood Tests for Indirect Influence in Private Reasoning Channels
di: Lorup, Alexander Boesgaard
Pubblicazione: (2026)

Cognitive Load Limits in Large Language Models: Benchmarking Multi-Hop Reasoning
di: Adapala, Sai Teja Reddy
Pubblicazione: (2025)

Harnessing Negative Signals: Reinforcement Distillation from Teacher Data for LLM Reasoning
di: Xu, Shuyao, et al.
Pubblicazione: (2025)

The Deterministic Horizon: When Extended Reasoning Fails and Tool Delegation Becomes Necessary
di: Guo, Dongxin, et al.
Pubblicazione: (2026)

DPO Unchained: Your Training Algorithm is Secretly Disentangled in Human Choice Theory
di: Zhou, Wenxuan, et al.
Pubblicazione: (2025)

Generalizing Numerical Reasoning in Table Data through Operation Sketches and Self-Supervised Learning
di: Cho, Hanjun, et al.
Pubblicazione: (2026)

Painless Activation Steering: An Automated, Lightweight Approach for Post-Training Large Language Models
di: Cui, Sasha, et al.
Pubblicazione: (2025)

The Instability of Safety: How Random Seeds and Temperature Expose Inconsistent LLM Refusal Behavior
di: Larsen, Erik
Pubblicazione: (2025)

Grokking in the Wild: Data Augmentation for Real-World Multi-Hop Reasoning with Transformers
di: Abramov, Roman, et al.
Pubblicazione: (2025)

Self-Training Doesn't Flatten Language -- It Restructures It: Surface Markers Amplify While Deep Syntax Dies
di: Liu, Ming
Pubblicazione: (2026)

Evaluating the Systematic Reasoning Abilities of Large Language Models through Graph Coloring
di: Heyman, Alex, et al.
Pubblicazione: (2025)

On the Limits of Learned Importance Scoring for KV Cache Compression
di: Steele, Brady
Pubblicazione: (2026)