官方发布 · 自媒体拆解
分类页是滚动库存，每页 24 条。首页只放当天精选。
On-premise assistants can give factory workers conversational access to machine documentation, but models capable of the task rarely fit shop-floor hardware. We show that after str
上下文子图做归纳关系学习。
势能引导的策略优化，打多轮 agent 任务。
仓库级规格编译，求忠实的 paper-to-code。
多模态 agent 做自动视频干预，强调证据。
Competitive programming has become a key test of large language model reasoning, with international competitions such as IOI and ICPC representing its most challenging settings. We
Article URL: https://arxiv.org/abs/2608.28482 Comments URL: https://news.ycombinator.com/item?id=49540787 Points: 3 # Comments: 0
表示对齐的潜在流匹配，做 SAR→EO。
把 GitHub 仓库压成可复用的 AI4AI skill。
计算匹配下的 MoE looped transformer 缩放。
The performance of LLM-based agents is jointly shaped by the base model and the harness used when interacting with the environment. This exposes them to safety risks in both harmfu
按程序家族巩固技能，自改进长程 agent。
用 LLM 训学生模拟器。
Harnessing naturally occurring feedback from user interactions offers a promising learning signal for Large Language Models (LLMs). However, recent studies suggest this feedback is
For more than 20 years, the Model-RB benchmark frb100-40 remained an open challenge; since 2014, its public record had stood at 99 of 100 variables. We give a directly checkable 10
mid-training 知识蒸馏，推理涨、事实记不住。
讨论像素级文本表示该怎么设计。
临床场景里人插手会让多 Agent 医疗系统崩。
竞赛编码后训练，冲金牌级表现。
稳定提示优化，把控制流和数据流拆开。
把 MLLM 当通用视觉-语言-动作智能体飞无人机。
propose/learn 循环，助手要可评估。
生成式检索器和强化学习一起长。
专家对照：dexterous 模仿学习的时间稳健性。