Image de couverture de livre

Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Zhao, Rosie, Meterez, Alexandru, Kakade, Sham, Pehlevan, Cengiz, Jelassi, Samy, Malach, Eran
Format:	Preprint
Publié:	2025
Sujets:	Machine Learning I.2.7
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Random Scaling of Emergent Capabilities
par: Zhao, Rosie, et autres
Publié: (2025)

On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMs
par: Zhao, Rosie, et autres
Publié: (2026)

Anytime Pretraining: Horizon-Free Learning-Rate Schedules with Weight Averaging
par: Meterez, Alexandru, et autres
Publié: (2026)

Repeat After Me: Transformers are Better than State Space Models at Copying
par: Jelassi, Samy, et autres
Publié: (2024)

Evaluating Relational Reasoning in LLMs with REL
par: Fesser, Lukas, et autres
Publié: (2026)

Universal Length Generalization with Turing Programs
par: Hou, Kaiying, et autres
Publié: (2024)

Generalizable LLM Learning of Graph Synthetic Data with Post-training Alignment
par: Zhang, Yizhuo, et autres
Publié: (2025)

The Role of Sparsity for Length Generalization in Transformers
par: Golowich, Noah, et autres
Publié: (2025)

WSM: Decay-Free Learning Rate Schedule via Checkpoint Merging for LLM Pre-training
par: Tian, Changxin, et autres
Publié: (2025)

Influence-driven Curriculum Learning for Pre-training on Limited Data
par: Schoenegger, Loris, et autres
Publié: (2025)

Post Hoc Extraction of Pareto Fronts for Continuous Control
par: Thakar, Raghav, et autres
Publié: (2026)

Evaluating the Efficacy of Hybrid Deep Learning Models in Distinguishing AI-Generated Text
par: Oketunji, Abiodun Finbarrs
Publié: (2023)

LoRA Soups: Merging LoRAs for Practical Skill Composition Tasks
par: Prabhakar, Akshara, et autres
Publié: (2024)

The Recurrent Transformer: Greater Effective Depth and Efficient Decoding
par: Oncescu, Costin-Andrei, et autres
Publié: (2026)

Seesaw: Accelerating Training by Balancing Learning Rate and Batch Size Scheduling
par: Meterez, Alexandru, et autres
Publié: (2025)

SeLeRoSa: Sentence-Level Romanian Satire Detection Dataset
par: Smădu, Răzvan-Alexandru, et autres
Publié: (2025)

Merge-Bench: Resolve Merge Conflicts with Large Language Models
par: Schesch, Benedikt, et autres
Publié: (2026)

Your Pretrained Model Tells the Difficulty Itself: A Self-Adaptive Curriculum Learning Paradigm for Natural Language Understanding
par: Feng, Qi, et autres
Publié: (2025)

Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs
par: Ahmadian, Arash, et autres
Publié: (2024)

The Data Efficiency Frontier of Financial Foundation Models: Scaling Laws from Continued Pretraining
par: Ponnock, Jesse
Publié: (2025)

Random Heterogeneous Neurochaos Learning Architecture for Data Classification
par: S, Remya Ajai A, et autres
Publié: (2024)

Scalify: scale propagation for efficient low-precision LLM training
par: Balança, Paul, et autres
Publié: (2024)

A Simplified Analysis of SGD for Linear Regression with Weight Averaging
par: Meterez, Alexandru, et autres
Publié: (2025)

ConciseRL: Conciseness-Guided Reinforcement Learning for Efficient Reasoning Models
par: Dumitru, Razvan-Gabriel, et autres
Publié: (2025)

Weight Pruning Amplifies Bias: A Multi-Method Study of Compressed LLMs for Edge AI
par: Rath, Plawan Kumar, et autres
Publié: (2026)

Learned Relay Representations for Forward-Thinking Discrete Diffusion Models
par: Rozonoyer, Benjamin, et autres
Publié: (2026)

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

FlexQuant: A Flexible and Efficient Dynamic Precision Switching Framework for LLM Quantization
par: Liu, Fangxin, et autres
Publié: (2025)

Scalable GPU-Accelerated Euler Characteristic Curves: Optimization and Differentiable Learning for PyTorch
par: Saxena, Udit
Publié: (2025)

Pre-trained Models Perform the Best When Token Distributions Follow Zipf's Law
par: He, Yanjin, et autres
Publié: (2025)

MALoRA: Mixture of Asymmetric Low-Rank Adaptation for Enhanced Multi-Task Learning
par: Wang, Xujia, et autres
Publié: (2024)

LLM Unlearning on Noisy Forget Sets: A Study of Incomplete, Rewritten, and Watermarked Data
par: Wang, Changsheng, et autres
Publié: (2025)

Procedural Environment Generation for Tool-Use Agents
par: Sullivan, Michael, et autres
Publié: (2025)

EvilGenie: A Reward Hacking Benchmark
par: Gabor, Jonathan, et autres
Publié: (2025)

On the Challenges of Creating Datasets for Analyzing Commercial Sex Advertisements to Assess Human Trafficking Risk and Organized Activity
par: Rivas, Pablo, et autres
Publié: (2024)

Unsupervised Evaluation of Multi-Turn Objective-Driven Interactions
par: Soroka, Emi, et autres
Publié: (2025)

Hierarchical Shift Mixing -- Beyond Dense Attention in Transformers
par: Forchheimer, Robert
Publié: (2026)

Spectral Clustering in Convex and Constrained Settings
par: Behera, Swarup Ranjan, et autres
Publié: (2024)

Mr. Snuffleupagus at SemEval-2025 Task 4: Unlearning Factual Knowledge from LLMs Using Adaptive RMU
par: Dosajh, Arjun, et autres
Publié: (2025)

Guided Speculative Inference for Efficient Test-Time Alignment of LLMs
par: Geuter, Jonathan, et autres
Publié: (2025)