共计 117 篇文章
2026
Why Steering Works：Toward a Unified View of Language Model Parameter Dynamics
LLM-VA： Resolving the Jailbreak-Overrefusal Trade-off via Vector Alignment
AlphaSteer： Learning Refusal Steering with Principled Null-Space Constraint
DiffuGuard： How Intrinsic Safety is Lost and Found in Diffusion Large Language Models
One-shot Optimized Steering Vector for Hallucination Mitigation for VLMs
RelayLLM： Efficient Reasoning via Collaborative Decoding
Text-to-LoRA： Instant Transformer Adaption
2025
A Unified Definition of Hallucination, Or： It's the World Model, Stupid
Blink：用于增强多模态理解的动态视觉token分辨率
JustRL： Scaling a 1.5B LLM with a Simple RL Recipe