React官方紧急公告：React Server Components存在未认证远程代码执行漏洞（CVSS 10.0），影响多个版本，请立即升级！
There is an unauthenticated remote code execution vulnerability in React Server Components. A fix has been published in versions 19.0.1, 19.1.2, and 19.2.1. We recommend upgrading immediately.
揭秘SGD在LLM预训练中不如Adam的根源：大有效学习率的关键作用。
arXiv:2605.17787v1 Announce Type: new Abstract: It is widely believed that stochastic gradient descent (SGD) performs significantly worse than adaptive optimizers such as Adam in pre-training Large La…
突破传统MMSE，用ProxiMAP提升PnP图像恢复性能，理论创新与实验验证兼具。
arXiv:2605.16396v1 Announce Type: cross Abstract: Plug-and-Play (PnP) methods have become standard tools for solving imaging inverse problems by replacing the intractable maximum a posteriori (MAP) de…
突破性研究：首次实现无限深和宽Transformer的可训练性，彻底解决深层网络训练瓶颈
arXiv:2605.17660v1 Announce Type: cross Abstract: Transformers have become the dominant architecture in modern machine learning, yet the theoretical understanding of their training dynamics remains li…
从形式语言学习视角，剖析微调与上下文学习的根本差异，ACL 2026 重磅论文。
arXiv:2604.23267v2 Announce Type: replace-cross Abstract: Large language models (LLMs) operate in two fundamental learning modes - fine-tuning (FT) and in-context learning (ICL) - raising key question…
聚焦实时流VLA架构创新，重新思考并加速Flow VLA推理效率，适合AI研究者。
arXiv:2603.19199v3 Announce Type: replace-cross Abstract: Real-time execution is crucial for deploying Vision-Language-Action (VLA) models in the physical world. Existing asynchronous inference method…
OpenAI对比预训练方法，学习文本与代码的高质量嵌入表示
Next.js 12 重磅发布：Rust 编译器提速5倍，引入中间件和React 18原生支持
Next.js 12 introduces a brand-new Rust compiler, Middleware (beta), React 18 Support, Native ESM Support, URL Imports, React Server Components (alpha), and more!
React Compiler 1.0 稳定版发布，自动记忆化优化组件，无需重写，已与主流工具链集成，生产就绪。
We are releasing the compiler's first stable release today.
React编译器已用于Instagram生产，即将开源，性能优化新突破。
In React Labs posts, we write about projects in active research and development. We’ve made significant progress since our last update, and we’d like to share our progress.
React核心团队官宣React 18发布计划、工作组及Alpha版本，新特性主打渐进式采用，开发者不容错过的版本更新解读
The React team is excited to share a few updates. We’ve started work on the React 18 release, which will be our next major version. We’ve created a Working Group to prepare the community for gradual a…
提出随机惩罚-障碍方法，解决约束机器学习优化难题，理论创新与算法实践兼备。
arXiv:2605.18618v1 Announce Type: new Abstract: Constrained machine learning enables fairness-aware training, physics-informed neural networks, and integration of symbolic domain knowledge into statis…
自动生成知识组件，让编程知识追踪更透明，从ACL论文看教育AI新突破。
arXiv:2502.18632v4 Announce Type: replace-cross Abstract: Knowledge components (KCs) mapped to problems help model student learning, tracking their mastery levels on fine-grained skills thereby facili…
揭秘GPU计算的黑箱，在CPU上精确复现NVIDIA矩阵乘法，解决机器学习非确定性难题。
arXiv:2603.20421v2 Announce Type: replace-cross Abstract: We present Hawkeye, a system for analyzing and reproducing GPU-level arithmetic operations. Using our framework, anyone can re-execute on a CP…
ICML 2026 录用，提出预测性预取策略加速检索增强生成，有效降低推理延迟。
arXiv:2605.17989v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) improves factual grounding in large language models but suffers from substantial latency due to synchronous retriev…
空气DNA采样技术革新生物多样性监测，Nature揭示关键方法细节与前沿应用
Nature, Published online: 19 May 2026; doi:10.1038/d41586-026-01604-7 Airborne DNA can yield insights with the right techniques
首个针对量化回测的LLM基准测试，被KDD 2026录用，评估自动策略生成能力。
arXiv:2605.17937v1 Announce Type: new Abstract: Quantitative backtesting is essential for evaluating trading strategies but remains hampered by high technical barriers and limited scalability. While L…
OpenAI发现梯度噪声尺度可预测神经网络训练并行性，为大规模训练提供理论基础。
We’ve discovered that the gradient noise scale, a simple statistical metric, predicts the parallelizability of neural network training on a wide range of tasks. Since complex tasks tend to have noisie…
首次定义具身动作表征的ImageNet基准，揭示人类视频数据可驱动机器人泛化学习。
LARYBench （Latent Action Representation Yielding Benchmark），一个指引从大规模的视觉数据学习到通用的隐式动作表征的系统化评测基准。实验结果表明：在动作泛化和控制精度上，通用视觉模型的表现均显著优于专门为具身智能设计的动作专家模型，具身动作表征可以从大规模人类视频数据中涌现。
用Rust自动化生成175种像素字体，彻底解决手动字距调整的繁琐与错误，步骤详解令人惊叹。
Here’s a fantastic post by Chevy Ray about how they made a dozens of pixel fonts using Rust : Kerning was a big time hog when making my previous font packs. Because the tools I was using required ever…
Figma如何重新定义行高与间距，揭秘Web设计史上精确控制与灵活性的博弈
Marcin Wichary has written yet another outstanding piece about how the team at Figma has redesigned how they treat line-height and spacing . But I can’t stop thinking about this bit where Marcin write…
批判当前AI多元对齐仅依赖偏好聚合，提出必须主动暴露分歧以实现真正价值多元主义。
arXiv:2605.14912v1 Announce Type: new Abstract: Pluralistic alignment is typically operationalised as preference aggregation: producing responses that span (Overton), steer toward (Steerable), or prop…
脉冲神经网络的局部学习规则综述与基准测试框架，助你快速理解不同训练算法的差异与适配场景
arXiv:2605.15058v1 Announce Type: cross Abstract: The rapid expansion of spiking neural networks (SNNs) has led to a proliferation of training algorithms that differ widely in biological inspiration, …
React Conf 2025大会上发布新组件<Activity/>、useEffectEvent以及React Native新特性，官方详细总结不容错过。
Last week we hosted React Conf 2025, in this post, we summarize the talks and announcements from the event...
React官宣Compiler Beta，自动记忆化优化性能，支持React 17+，并开放工作组供社区参与
At React Conf 2024, we announced the experimental release of React Compiler, a build-time tool that optimizes your React app through automatic memoization. In this post, we want to share what's next f…
React官方团队披露Server Components等最新研究进展，深度解读新一代应用架构。
In React Labs posts, we write about projects in active research and development. We've made significant progress on them since our last update, and we'd like to share what we learned.
React 的零包体积服务端组件来了，颠覆传统数据获取与渲染模式，一睹下一代 React 架构。
2020 has been a long year. As it comes to an end we wanted to share a special Holiday Update on our research into zero-bundle-size React Server Components.
非CUDA平台优化新突破：在Apple M3 Ultra上对扩散模型推理进行10阶段系统实验，揭示实时图像生成加速方案。
arXiv:2605.16259v1 Announce Type: new Abstract: While real-time image generation using diffusion models has advanced rapidly on NVIDIA GPUs, systematic optimization research on non-CUDA platforms such…
教你利用激活几何诊断转向瓶颈，GRACE框架让概念工程优化更高效。
arXiv:2605.16362v1 Announce Type: new Abstract: Activation steering offers a lightweight way to control LLMs without retraining, but its effectiveness varies sharply across concepts. Prior work often …
提出可区分删除方法，统一知识擦除与拒绝机制，为LLM去学习难题提供新思路。
arXiv:2605.16776v1 Announce Type: new Abstract: Mitigating sensitive and harmful outputs is fundamental to ensuring safe deployment of LLMs. Existing approaches typically follow two paradigms: Knowled…
用混合数字孪生+LSTM实现1型糖尿病长期血糖预测，精准度突破现有方法。
arXiv:2605.16860v1 Announce Type: new Abstract: Accurate long-horizon glucose forecasting is critical for automated insulin delivery systems, which help people with type 1 diabetes (T1D) manage their …
提出逐步评分奖励机制，优化LLM推理的中间步骤监督，突破传统仅奖励最终答案的局限。
arXiv:2605.17291v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning in large language models, but rewards only final-answer correc…
脑部MRI自监督学习新框架，利用3D体令牌对齐实现跨数据集通用表征
arXiv:2605.16775v1 Announce Type: cross Abstract: Self-supervised learning (SSL) has advanced medical image analysis be enabling learning form large unlabelled data. However, in brain magnetic resonan…
针对异构LLM池的混合动态路由架构，无需因模型目录变更而重训练，解决现有二元路由的耦合问题。
arXiv:2605.17106v1 Announce Type: cross Abstract: Production LLM deployments increasingly maintain heterogeneous model pools spanning order-of-magnitude cost differences. Existing routers make binary …
揭秘多语言下AI安全护栏失效的根本原因，从数据分布到tokenization的深度剖析。
arXiv:2605.17173v1 Announce Type: cross Abstract: Large language models exhibit safety degradation in non-English languages. Standard evaluation relies on Jailbreak Success Rate (JSR), which confounds…
提出无近似、无需训练的粒子滤波方法用于扩散代理模型，被ICML 2026接收，开创性极强。
arXiv:2605.18745v1 Announce Type: cross Abstract: Diffusion-based generative models increasingly rely on inference-time guidance, adding a drift term or reweighting mixture of experts, to improve samp…
如何自动将神经网络实现从一种深度学习框架迁移到另一种？这篇论文提出新方法，解决框架间兼容性难题
arXiv:2511.02610v2 Announce Type: replace Abstract: The development of smart systems (i.e., systems enhanced with AI components) has thrived thanks to the rapid advancements in neural networks (NNs). …
多模态大模型消除幻觉新方法：锐度感知鲁棒擦除，超越浅层遗忘，提升模型可靠性
arXiv:2601.16527v2 Announce Type: replace Abstract: Multimodal LLMs are powerful but prone to object hallucinations, which describe non-existent entities and harm reliability. While recent unlearning …
联邦学习下用合成表格数据自我优化大模型，Concordia框架有望提升隐私与效率。
arXiv:2605.09855v2 Announce Type: replace Abstract: Federated learning (FL) enables training large language models (LLMs) without sharing raw data, but adapting LLMs under strict data isolation and no…
长上下文LLM无需重训练即可轻松转为混合模型，LightTransfer方法实现高效适配，论文被TMLR 2025收录。
arXiv:2410.13846v3 Announce Type: replace-cross Abstract: Scaling language models to handle longer contexts introduces substantial memory challenges due to the growing cost of key-value (KV) caches. M…
RSS 2026 论文提出自监督方法，让机器人通过预测行动自主学习推理，无需大量标注数据。
arXiv:2602.08167v2 Announce Type: replace-cross Abstract: Embodied Chain-of-Thought (CoT) reasoning has significantly enhanced Vision-Language-Action (VLA) models, yet current methods rely on rigid te…
ICML 2026前沿成果，提出异常偏好图像生成新范式，为可控生成开辟新方向。
arXiv:2605.02439v2 Announce Type: replace-cross Abstract: Synthesizing realistic and diverse anomalous samples from limited data is vital for robust model generalization. However, existing methods str…
用非洲低资源语言玩多轮对话，成功绕过ChatGPT、Gemini等主流大模型的安全护栏，安全漏洞新发现。
arXiv:2605.18239v1 Announce Type: new Abstract: Large Language Models (LLMs) remain vulnerable to jailbreak attempts that circumvent safety guardrails. We investigate whether multi-turn conversations …
自主AI可信度关键：显式溯源机制保障责任伦理，前沿论文剖析Agentic AI治理新范式
arXiv:2605.17169v1 Announce Type: cross Abstract: Agentic AI is rapidly proliferating across diverse real-world domains such as software engineering, yet public trust has not kept pace. The central re…
MoE架构在严格等资源条件下首次证明超越稠密大模型，ICLR 2026最新研究。
arXiv:2506.12119v2 Announce Type: replace Abstract: Mixture-of-Experts (MoE) language models dramatically expand model capacity and achieve remarkable performance without increasing per-token compute.…
重磅研究：代码领域的缩放定律显示需要比自然语言多几个数量级的数据才能达到相同性能提升，引发对大模型训练数据效率的重新思考。
arXiv:2510.08702v2 Announce Type: replace Abstract: Code Large Language Models (LLMs) are revolutionizing software engineering. However, scaling laws that guide the efficient training are predominantl…
提出Evo-Memory基准，全面评估LLM Agent通过自演化记忆进行测试时学习的能力。
arXiv:2511.20857v2 Announce Type: replace Abstract: Statefulness is essential for large language model (LLM) agents to perform long-term planning and problem-solving. This makes memory a critical comp…
LLM极低比特量化新突破：SignRoundV2大幅缩小性能差距，实现高效训练后量化
arXiv:2512.04746v2 Announce Type: replace Abstract: Extremely low-bit quantization is critical for efficiently deploying Large Language Models (LLMs), yet it often leads to severe performance degradat…
AI对齐在边界场景下突然失效，揭示了现有安全方法的根本漏洞。
arXiv:2602.20042v2 Announce Type: replace Abstract: General Alignment has improved average-case helpfulness and safety, but current alignment practice still rewards confident, single-turn responses. T…
ACL 2025论文提出Sparse-to-Dense方法，实现视频理解在LLM中的无损加速，堪称"免费午餐
arXiv:2505.19155v2 Announce Type: replace-cross Abstract: Due to the auto-regressive nature of current video large language models (Video-LLMs), the inference latency increases as the input sequence l…