Recent Posts

Decoupling Exploration from Optimization in RLVR

解耦 RLVR 中的探索与优化：ExpDis 框架如何激发大模型推理多样性

10/8/2026

arxivaics.ai

EngramEdit: Decoupled Knowledge Updates in LLMs through Conditional Memory

EngramEdit：利用条件记忆解耦大语言模型的事实知识编辑

10/8/2026

arxivaics.cl

Agent in a Bottle: Can LLM Agents Turn Their Capabilities Into Cheap, Scalable Artifacts?

装瓶智能体：大模型能否将自身能力编译为廉价高效的复用制品？

10/7/2026

arxivaics.ai

World Models' Last Exam in Physics

世界模型的物理学终考：基于物理测量的严苛基准检验

10/7/2026

arxivaics.cv

Base Models Can Reason By Taking a Cue From Training Data

基座大模型本就会推理：前缀提示词如何激活预训练的思维潜能

10/6/2026

arxivaics.ai

CLIFT: Conformal Self-Verification for Web Agent Training and Test-Time Scaling

CLIFT：基于共形自验证的网页智能体训练与测试期扩展

10/6/2026

arxivaics.ai

LESSER: Post-Training Data Selection with Output-Layer Gradients

LESSER：用输出层梯度实现低成本大模型后训练数据选择

10/5/2026

arxivaics.lg

What Should World Models Forget? Stratified Retention for Continual Adaptation

世界模型应该遗忘什么？持续适应中的分层保留范式

10/5/2026

arxivaics.ai

On the Off-Policy Teacher in On-Policy Distillation

同策略蒸馏中的离策略教师困局

10/3/2026

arxivaics.ai

Semifactual Credit-Augmented Policy Optimization

半事实信用增强策略优化：瓦解 GRPO 的虚假关联

10/1/2026

arxivaics.ai

Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering

Imagine3D-LLM：让多模态大模型在作答前先构想 3D 场景

9/30/2026

arxivaics.cl

Telescopic Language Models

望远镜语言模型：单次训练贯通全深度算力连续谱

9/29/2026

arxivaics.ai