论文精选 83
Behavioral Couplings: From Formation to Behavioral Emergence in Language Models· 行为耦合：从形成到行为涌现
Fine-tuning large language models (LLMs) can induce behaviors beyond those explicitly targ…
领域：cs.CL作者：Dongyoon Hahm、Woogyeol Jin、Dennis Wei
📰 AI 研究简报 · 10月20日
今日头条：Meshy推出Meshy Edit：基于全新自研AI模型，一句指令实现3D模型局部编辑 —— 圈选修改区域，一句话描述改动，约一分钟生成新模型，未选中区域保持原样
共 54 条预计阅读 32 分钟Token 消耗 97,200
arXiv 最新论文
论文精选 60已解读
AI-Driven Reproducibility and Analytical Robustness in Scientific Discovery· AI 驱动的科学发现可复现性与分析鲁棒性
Reproducible and analytically robust research findings are essential to scientific progres…
领域：cs.AI作者：Shicheng Liu、Krista Opsahl-Ong、Andrew Drozdov
论文精选 82
DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses· DiscoveryRSI: 递归学习发现有价值假设
Scientific discovery requires developing worthwhile hypotheses into practical investigatio…
领域：cs.CL作者：Jianyuan Zhong、Chenwei Wu、Zhaochen Yu
论文精选 83
Beyond the Answer: Student-Conditioned Trace Construction for Reasoning Distillation· 超越答案：面向学生的推理蒸馏
A reasoning trace can accompany a correct answer without providing effective supervision f…
领域：cs.CL作者：Shaoang Li、Yanhang Shi、Yufei Li
论文精选 60已解读
Self-Building Labs· 自建实验室
Self-driving laboratories (SDLs) accelerate scientific discovery by automating experimenta…
领域：cs.AI作者：Jiaru Bai、Yuchi Zhao、Han Hao
论文精选 83
Acceptance Is Not Speed: Length Inflation in Training-Free Lossy Speculative Decoding· 接受并非速度：无训练损失性投机解码中的长度膨胀
Speculative decoding typically characterizes speedup by draft tokens accepted per verifica…
领域：cs.CL作者：Chiatzen Wang、Bill Xu、Haochen Gu
论文精选 82
Steering beyond Language: Improving Multilingual Training-Free Text Embeddings via Paired-Residual Subspace Removal· 超越语言：通过配对残差子空间移除改善多语言无训练文本嵌入
While deriving text embeddings directly from frozen LLMs has emerged as a promising direct…
领域：cs.CL作者：Ailiang Lin、Zhuoyun Li、Yusong Wang
论文精选 60已解读
Beyond SAEs: Consistent Concept Discovery with Sparse Coding· 超越 SAE：用稀疏编码一致发现概念
Sparse autoencoders (SAEs) explain foundation model representations by decomposing activat…
领域：cs.AI作者：Jinho Choi、Hyesu Lim、Jaegul Choo
论文精选 60已解读
FuTransformer: FP8 prefill and INT4 decode for Llama-style inference on Hopper
Transformer prefill reuses weights across tokens and is dominated by matrix multiplication…
领域：cs.AI作者：Wei Wang
论文精选 60已解读
DrugHarness: An Evolvable Agent Harness for Computational Drug Discovery· 药物发现的计算框架
Computational drug discovery requires agents to compose scientific operations and adapt to…
领域：cs.AI作者：Yuxing Chen、Zhisheng Zhang、Cheng Tang
论文精选 60已解读
Beyond Large Language Models: A Mental World Model Framework for Mental Modeling, Simulation, Evolution, and Interpretability· 超越大语言模型：心理世界模型框架
Understanding and simulating how subjective cognition and mental states emerge, evolve, an…
领域：cs.AI作者：Qingshi Wang
论文精选 60已解读
OwlSight: A Panoramic Monitor for Chain-of-Thought Reasoning· OwlSight：链式思考的全景监控
Chain-of-Thought (CoT) monitoring offers actionable evidence for oversight AI systems, wit…
领域：cs.AI作者：Lewen Yan、Yanxu Zhu、Jie Zhang
论文精选 60已解读
GEMS: Molecular Structure Elucidation as Learned Search over Isomer Graphs· GEMS：基于同分异构体图的分子结构解析
Identifying small-molecule structures from molecular formulas and tandem mass spectra rema…
领域：cs.AI作者：Utku Umur Acikalin、Yingheng Wang、Goncalo J. Gouveia
论文精选 60已解读
From Trajectories to Differentiated Skills: Cognitive Differentiation for Self-Improving Agents· 从轨迹到差异化技能：自我改进代理的认知分化
Large language model (LLM) agents tackle complex tasks through reasoning and interaction w…
领域：cs.AI作者：Yewei Zhao、Derong Xu、Wei Chen
论文精选 60已解读
RL-Guided Temporal Localization for Dual-Channel Retrieval in Long-Horizon Agent Memory· 长期视野智能体记忆的双通道检索时序定位
Time-aware retrieval is crucial for long-horizon LLM agents to effectively leverage ever-g…
领域：cs.AI作者：Dabin Sheng、Zhe Wu、Jinming Zhao
论文精选 60已解读
Darwin-Claw: Environment-Adaptive Experience Consolidation for Recursively Self-Evolving Agents· Darwin-Claw: 面向环境的体验整合
While Large Language Model (LLM) agents have demonstrated remarkable capabilities across c…
领域：cs.AI作者：Mingze Yin、Xiaohan Wang、Dian Li
论文精选 60已解读
SkillBrowser: Selective Skill Exposure for Scalable Skill-Using Agents· SkillBrowser：选择性技能暴露
Large language model agents increasingly rely on skills to extend their capabilities in co…
领域：cs.AI作者：Jin Ye、Yifan Jiang、Lan-zhe Guo
论文精选 82
Trapped in Your Story: Evaluating Narrative-Induced Sycophancy in Large Language Models· 困于你的故事：评估叙事影响下的马屁现象
Large language models (LLMs) are known to exhibit sycophancy by adjusting their responses …
领域：cs.AI作者：Runtai Yang、Dandan Zhu、Kaiwei Zhang
论文精选 83
LocalIntent: Outcome-Specific Structured Generation for Backend-Compatible Smart Building Automation with Local LLMs· 本地大模型用于智能建筑自动化
Locally deployed large language models (LLMs) can reduce reliance on remote cloud services…
领域：cs.AI作者：Tianyang Liu、Jiayue Wang、Zhanhao Cheng
论文精选 83
S-GAP: Stealthy Graph Attack Primitives for Graph Large Language Models· S-GAP: 针对图大语言模型的隐蔽性攻击原语
Graph Large Language Models (GraphLLMs) enable powerful semantic understanding and reasoni…
领域：cs.AI作者：Wenda Li、Tongya Zheng、Shunyu Liu
论文精选 83
GatewayBench: Unified Capability and Security Evaluation Platform for MCP Gateways· GatewayBench: 统一的 MCP 网关能力与安全评估平台
AI agents have increasingly interacted with external environments by invoking tools, with …
领域：cs.AI作者：Yitian Zheng、Zhaorun Chen、Haoyu Li
论文精选 83
ScoutRoute: Efficient Mixture-of-Experts Routing Using Internal Language Model Signals· ScoutRoute：利用语言模型信号的高效 MoE 路由
Efficient expert routing is a fundamental challenge in Mixture-of-Experts (MoE) systems, p…
领域：cs.AI作者：Rajdeep Pandhare、Delzaan Bharucha
论文精选 82
Generative Policy Improvement from Observational Data· 从观测数据生成改进策略
We study how to improve decision policies from observational data to generate better actio…
领域：cs.AI作者：Jonas Schweisthal、Dennis Frauen、Stefan Feuerriegel
论文精选 85
Monitor-Validity Failures: LLM Safety Evidence Paths at Evolving Framework Boundaries· 大模型安全监测的框架边界问题
Modern LLM applications rely on runtime monitors that inspect tool calls, traces, streams,…
领域：cs.AI作者：Junchi Liao
论文精选 82
OliO: ODE-based Linear Transition Operator for Self-Supervised Time Series Forecasting· OliO: 基于ODE的线性转换算子
Self-supervised representation learning has advanced time series forecasting by capturing …
领域：cs.AI作者：Kwangryeol Park、Seulki Lee
GitHub 热门开源项目
开源项目精选 83
AndreSzymanski15/ai-jailbreaks· AI越狱提示
AI jailbreaks: 35 paste-ready AI jailbreak prompts for 25+ frontier and open-weight LLMs -…
⭐ 254 stars
开源项目精选 82
SubmitGame/awesome-ai-game-dev-tools· 优秀 AI 游戏开发工具合集
Open-source AI tools for making games, and for decompiling, remixing and mashing up games …
⭐ 40 stars语言 Python
精选 AI 行业资讯
行业资讯精选 80
US orders federal agencies to remove China’s DeepSeek AI from their systems· 美国下令联邦机构移除中国的 DeepSeek AI
The US has ordered federal civilian agencies to remove DeepSeek products and third-party s…
行业资讯精选 80
Eka is worth $1.45B thanks to a physical AI model that has experts raving· Eka：物理 AI 模型使专家赞叹
Eka Robotics may be the first robotic arm that is actually good at picking a wide range of…
行业资讯精选 80
Yelp can now call on your behalf to check wait times at a restaurant· Yelp 推出新 AI 功能
Yelp is embedding numerous new AI features into its app that can do everything from call t…
行业资讯精选 70
Otter.ai raises $320 million from General Catalyst’s growth fund for its meeting tool· Otter.ai 获 3.2 亿美元融资
Meeting note-taking tool Otter.ai announced on Tuesday that it has raised $320 million fro…
行业资讯精选 82
Spain-Based Startups Advance Regional Innovation With Barcelona Supercomputing Center· 西班牙初创公司利用巴塞罗那超算中心推进区域创新
The Barcelona Supercomputing Center (BSC)’s AI Factory Accelerator Programme, now in full …
行业资讯精选 85
Meshy推出Meshy Edit：基于全新自研AI模型，一句指令实现3D模型局部编辑
圈选修改区域，一句话描述改动，约一分钟生成新模型，未选中区域保持原样
行业资讯精选 80
AI硬件出海，别当“背景板”，GTC2026 Shanghai「AI硬件专区」邀你占位
2026 年 11 月 4 日至 5 日，GTC2026 全球流量大会（上海）将在上海世博展览馆 H3 馆举办
行业资讯精选 80
中科通量亮相2026 RISC-V中国峰会，全栈RISC-V智能覆盖视频+智算全场景
10月18日，第六届RISC-V中国峰会在深圳会展中心（福田）开幕。中科通量携SmarCo AI工作站、SmarCo智能视频编码一体机、SmarCo智算加速卡等产品与解决方案亮相，…
行业资讯精选 62
Mamdani says feds have abdicated on AI, urges New Yorkers to push back· 纽约市长呼吁抵制 AI 叙事并要求监管
Mayor Zohran Mamdani said Monday night that New Yorkers must resist the narrative that an …
行业资讯精选 65
Claude新模型曝光！能跳出思维定势，修Bug治标又治本
OpenAI只掏出一个6.1 Terra…
行业资讯精选 85
Nscale and Anthropic are reportedly in talks for yet another $45 billion compute deal· Nscale 和 Anthropic 再度商讨450亿美元计算交易
In August, Nscale and Anthropic announced a $45 billion compute deal. The two companies ar…
行业资讯精选 75
Embedded AI evaluators won’t be an easy solution, Hugging Face’s policy chief says· 嵌入式AI评估并非易事
AI labs have staked their reputations on model benchmarking — the supposedly neutral and a…
行业资讯精选 82
Microsoft says AI agents are like insider threats as ‘frontier models can’t be trusted’· 微软警告：AI代理如同内部威胁
Microsoft wants enterprises to closely watch the agents employees use with systems not con…
行业资讯精选 82
Amazon is taking its biggest shot at Nvidia yet with new TorchNeuron dev tool· 亚马逊推出新 TorchNeuron 开发工具挑战英伟达
Amazon wants to break Nvidia’s hold on PyTorch, the free toolkit used to build most AI app…
行业资讯精选 83
Hackers hit AI coding tool OpenCode, running up charges on users’ AI accounts
OpenCode is an open-source AI coding agent that's become a popular alternative to Claude C…
行业资讯精选 82
Germany Wants to Lead the World in Industrial AI: How NVIDIA Will Help· 德国想要领导全球工业 AI：NVIDIA 如何助力
In the cavernous hall of Kraftwerk, a former power station in Berlin, German Chancellor Fr…
行业资讯精选 82
Tolan snags $30M for its AI companion that helps users navigate life· Tolan 获 3000 万美元融资，助用户导航生活
AI companion app Tolan has raised another $30 million after attracting more than 250,000 a…
行业资讯精选 65
Replika’s founder built the AI assistant she pitched at Disrupt 12 years ago· Replika创始人推出新AI助手Wabi
Eugenia Kuyda's newest startup Wabi is positioned to take advantage of an idea she had mor…
行业资讯精选 82
Tenstorrent hits $6.2B valuation to scale open-weight AI chips· Tenstorrent 获 15 亿美元融资，加速开放权重 AI 芯片发展
After drawing acquisition interest earlier this year, the Nvidia challenger raised $1.5 bi…
行业资讯精选 65
People are using AI to make their smart TVs dumb again· 用 AI 让智能电视变简单
Fed up with ads, tracking and sluggish interfaces, some smart TV owners are turning to AI …
行业资讯精选 82
Nvidia launches new open models to run agents in the enterprise and on PCs· Nvidia 发布新开放模型，推动企业级 AI 代理
Nvidia is leaning hard into the trend of enterprise companies wanting to run their own ope…
行业资讯精选 80
NVIDIA Advances Agentic Security for Media, From European Public Service Broadcasting to Global Content Production· NVIDIA 推动智能媒体安全
At a creative industry event that took place today in Berlin — hosted in collaboration wit…
行业资讯精选 85
通研院孵化的团队，也盯上了RSI
当AI越来越懂你，如何让智能真正属于你？
行业资讯精选 82
Advancing coding in low‑resource programming languages· 在低资源编程语言中推进编码
Learn how OpenAI is improving ChatGPT across low-resource programming languages, with new …
YouTube 热门 AI 视频
AI 视频精选 80
Learn to build to better AI Agents through better Context· 通过改进上下文构建更好的AI代理
YouTube · DeepLearningAI
📺 DeepLearningAI▶ 74 次观看
AI 视频精选 75
Cooking with Codex: Ep. 1· 用 Codex 编程：第 1 集
YouTube · OpenAI
📺 OpenAI▶ 817 次观看
AI 视频精选 65
AI factories need to respond when the power grid needs relief.· AI 工厂需响应电网压力
YouTube · NVIDIA
📺 NVIDIA▶ 236 次观看