William Yang Wang

LogicReward: Incentivizing LLM Reasoning via Step-Wise Logical Supervision

DevOps-Gym: Benchmarking AI Agents in Software DevOps Cycle

Dynamic Speculative Agent Planning

SOPBench: Evaluating Language Agents at Following Standard Operating Procedures and Constraints

Adversarial Training for Process Reward Models

Do Larger Language Models Generalize Better? A Scaling Law for Implicit Reasoning at Pretraining Time

Self-Resource Allocation in Multi-Agent LLM Systems

PromptArmor: An Essential Baseline for Prompt Injection Defenses

Cost-effective Agent Test-time Scaling via Budget-Aware Thinking

HexMachina: Self-Evolving Multi-Agent System for Continual Learning of Catan

MuSLR: Multimodal Symbolic Logical Reasoning

Weak-to-Strong Jailbreaking on Large Language Models

ThoughtTerminator: Benchmarking, Calibrating, and Mitigating Overthinking in Reasoning Models

Speculative Knowledge Distillation: Bridging the Teacher-Student Gap Through Interleaved Sampling

T2V-Turbo-v2: Enhancing Video Model Post-Training through Data, Reward, and Conditional Guidance Design

MMWorld: Towards Multi-discipline Multi-faceted World Model Evaluation in Videos

MMSci: A Dataset for Graduate-Level Multi-Discipline Multimodal Scientific Understanding

COrAL: Order-Agnostic Language Modeling for Efficient Iterative Refinement

MLGym: A New Framework and Benchmark for Advancing AI Research Agents

Gödel Agent: A Self-Referential Framework Helps for Recursively Self-Improvement

Weak-to-Strong Jailbreaking on Large Language Models

Discovering Factor Level Preferences to Improve Human-Model Alignment

Understanding the Interplay between Parametric and Contextual Knowledge for Large Language Models

Generalization v.s. Memorization: Tracing Language Models’ Capabilities Back to Pretraining Data

VSP: Assessing the dual challenges of perception and reasoning in spatial planning tasks for MLLMs

TC-Bench: Benchmarking Temporal Compositionality in Conditional Video Generation

Can Editing LLMs Inject Harm?

MELON: Provable Defense Against Indirect Prompt Injection Attacks in AI Agents

Compact Multimodal Context Represenations Using Visual Tokens

Pixelated Instructions: Can Multimodal Large Language Models Follow Printed Instructions in Images?

SWE-Search: Enhancing Software Agents with Monte Carlo Tree Search and Iterative Refinement

Detecting Training Data of Large Language Models via Expectation Maximization

DebUnc: Improving Large Language Model Agent Communication Via Uncertainty Metrics