论文精选 65已解读
On the estimation and validity of AI time horizons---a statistical look at the METR plot· AI时间预测的统计分析
METR's 50\% time horizon measures the human completion time of software tasks that an AI s…
领域：cs.AI作者：Drew T. Nguyen、William Fithian
📰 AI 研究简报 · 10月10日
今日头条：JEPA、空间智能路线正面交锋：WorldArena 2.0 挑战赛开考「能不能被机器人真正用起来」| IROS 2026 —— <img data-src="https://static.leiphone.com/uploads/new/images/20261010/6ac9b2326d2ea.jpg" class="rich_pages wxw-img" data-ratio="0.55" data-s="300,640" data-type="jpeg" data-w="1000" style="width:100%;display:inline-block;text-align:center;ba
共 53 条预计阅读 32 分钟Token 消耗 95,400
arXiv 最新论文
论文精选 65已解读
BrickBench: Evaluating Agentic Brick Design· BrickBench: 代理LEGO设计评估
We propose BrickBench, a benchmark for agentic text-conditioned LEGO-set design. Given a p…
领域：cs.AI作者：Peter Kulits、Yiqing Xu、R. Kenny Jones
论文精选 65已解读
Ecology of AI Agents: Collaboration Creates a Population Threshold for Takeoff· AI 代理的生态：协作创造起飞的人口门槛
AI agents can now conduct real-world cyberattacks, scale up capabilities with the number o…
领域：cs.AI作者：Erin Crawley、Hidenori Tanaka
论文精选 82
Predicting Alignment Generalization with Value Representations· 用价值表示预测对齐泛化
LLM developers post-train their models to exhibit prosocial values and behavioral traits, …
领域：cs.CL作者：Andy Liu、Mehar Bhatia、Karolina Stanczak
论文精选 65已解读
Searching for "Harmful Refusal": A Psychometric Audit of an AI Safety Benchmark
Safety benchmarks typically report one overall score for a suite of datasets, each of whic…
领域：cs.AI作者：Christopher M. Stewart、Preston Botter、Natalie Sarabosing
论文精选 65已解读
HRIL: Learning Multimodal Synergy via Higher-Order Tensor Modeling· HRIL：通过高阶张量建模学习多模态协同
Self-supervised multimodal representation learning has achieved remarkable success across …
领域：cs.AI作者：Qun Dai、Liangjian Wen、Jiang Duan
论文精选 65已解读
GeoReform: Reflective Formalization Evolution for Multimodal Geometry Problem Solving· GeoReform：多模态几何问题解决的反思式形式化进化
Multimodal large language models (MLLMs) often struggle to identify and use geometric rela…
领域：cs.AI作者：Jialu Wang、Ruichen Zhang、Xiaoou Liu
论文精选 82
Latent Core Tokenizer: Compress, but Meaningfully· 潜在核心分词器：有意义的压缩
Tokenizers are commonly optimized for compression, but a compact vocabulary does not neces…
领域：cs.CL作者：Felermino D. M. A. Ali、Millicent Ochieng、Ogbemi Ekwejunor-Etchie
论文精选 65已解读
OnTrack: Real-Time Monitoring and Intervention in LLM Agent Trajectories via Streaming Structure-Aware Optimal Transport· OnTrack：大模型代理路径实时监控与干预
Agents are deployed in applications from trip planners and stock trading to IT incident tr…
领域：cs.AI作者：Babak Barazandeh、Connor Swanson、Chinmay Kulkarni
论文精选 83
Which Skill to Distill? SGUID: Selecting a Compact Skill Bank for Model-Skill Co-Evolution· 选择模型技能共演化所需的紧凑技能库
Skills, reusable procedural guidance added at inference, can substantially improve LLM dow…
领域：cs.CL作者：Yuhan Liu、Xiyao Ma、Zhongkai Sun
论文精选 65已解读
HANS: A Handwritten Answer Sheet Dataset for Noisy Hybrid Document Parsing· HANS: 手写答题纸数据集
Intelligent grading and automated scoring technologies constitute critical infrastructure …
领域：cs.AI作者：Xiazhen Wu、Wansong Qin、Yangbin Zheng
论文精选 65已解读
Cited but Not Consulted: A Counterfactual Audit of Legal Chain-of-Thought Faithfulness
Large language models increasingly justify legal decisions by naming the statute or preced…
领域：cs.AI作者：Saisab Sadhu、Shreeyans Arora、Pratinav Seth
论文精选 65已解读
Accurate but Not Humble: Evaluating Epistemic Humility in LLM Agents under Knowledge Conflict· 准确但不谦逊：评估大模型代理的认知谦逊
When retrieved evidence contradicts an agent's prior beliefs, does it revise its answer, a…
领域：cs.AI作者：Kaiser Sun、Bernal Jimenez Gutierrez、Hongjun Liu
论文精选 65已解读
Overcoming Prior Barriers: Supervised Fine-Tuning under Long-Tail Distribution· 克服先验障碍：长尾分布下的监督微调
Supervised fine-tuning (SFT) adapts pretrained large language models (LLMs) to downstream …
领域：cs.AI作者：Haohui Wang、Jiahao Xu、Wangzhi Zhan
论文精选 65已解读
Can AI Agents Learn Their Way to the Top? Evaluating Heuristic Learning in a Long-Running Game Agent Competition· AI 智能体能否进步到顶峰？
Adversarial games have driven advances from heuristic search to reinforcement learning, ye…
领域：cs.AI作者：Kaisen Yang、Qingle Liu、Kejin Wang
论文精选 83
VFold: Symmetry-Aware Cross-Layer Value Cache Compression· VFold: 跨层值缓存压缩
While caching key-value (KV) states accelerates Large Language Model (LLM) decoding, this …
领域：cs.CL作者：Neha Verma、Sungwon Kim、Kenton Murray
论文精选 82
Prior or Feedback? What an LLM Uses When Adapting Neural Operators· 先验还是反馈？LLM在适应神经算子时的使用
Do LLM scientific agents rely only on their initial task context, or do they adapt their d…
领域：cs.AI作者：Julian Chan、Javier Mora Jimenez
论文精选 82
Verdict Without the Rule: Diagnosing and Auditing Regulatory Rule Sensitivity in LLM Compliance Systems· 无规则的判决：评估LLM合规系统的规则敏感性
Large language model compliance systems are deployed on the assumption that a verdict depe…
领域：cs.AI作者：Saisab Sadhu、Aadit Sengupta、Vinay kumar Sankarapu
论文精选 85
A Structural Theory of Cognitive Representation and Problem Solving,Contexts, Invariance, and the Knowledge Space· 认知表示与问题解决的结构理论
Learning and problem solving depend critically on the structure of internal representation…
领域：cs.AI作者：Antal Jakovác、András Telcs
论文精选 85
Looking Inside LLMs: Small-World Connectivity as a Signature of Reasoning Performance· 探究LLM推理性能的小世界网络特性
Understanding large language model (LLM) reasoning requires looking beyond behavioral perf…
领域：cs.AI作者：Zheng Huang、Sansheng Cao、Enpei Zhang
论文精选 85
Learning Probabilistic Logic Programs with Functional Gradient Guided Language Models· 使用梯度引导语言模型学习概率逻辑程序
Declarative logic programs offer a powerful and interpretable abstraction for encoding rel…
领域：cs.AI作者：Saurabh Mathur、Sahil Sidheekh、Bhavan Vasu
论文精选 83
One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails· 单词开闸：决策模型防护攻击
A typed decision model reads a piece of text and returns a probability over caller-defined…
领域：cs.AI作者：Seyedarmin Azizi、Erfan Baghaei Potraghloo、Massoud Pedram
论文精选 85
HarnessSQL: Harness-Native Training for SQL Agents in Realistic Database Environments· HarnessSQL: SQL 代理的本机训练框架
Text-to-SQL models are commonly trained to map questions directly to static queries, where…
领域：cs.CL作者：Haolin Yang、Jipeng Zhang、Jian Xie
论文精选 83
When Has a Bayesian Neural Network Sampled Enough? Adaptive Inference Time with Statistical Guarantees· 贝叶斯神经网络的自适应推理时间
Bayesian neural network predictions are commonly approximated using a fixed number of Mont…
领域：cs.AI作者：Fabian Denoodt、Sibylle Hess
论文精选 82
Recursive Self-Improvement through Multi-Agent Self-Supervision· 通过多智能体自监督实现递归自我改进
Recursive self-improvement (RSI) of a model on non-verifiable tasks, such as open-ended re…
领域：cs.AI作者：Hyunin Lee、Jinglue Xu、Jeffrey Seely
GitHub 热门开源项目
开源项目精选 82
Microqian2th/dsh-codex-effort-slider· Codex 风格推理等级滑条
Codex 风格的推理等级滑条（DeepSeek Harness 插件）：可拖动连续滑条 + 高档位渐变星尘特效
⭐ 61 stars语言 JavaScript
开源项目精选 85
Folkbench/folkbench
Folkbench 是一个可信的 AI 中转站评测与服务目录，按模型比较可用率、响应速度、价格、隐私与运营证据
⭐ 158 stars语言 TypeScript
开源项目精选 75
thesnarkitecht/rembrandt
A free photo editor. No account, no subscription, no tracking. RAW, masks and on-device AI…
⭐ 86 stars语言 C++
精选 AI 行业资讯
行业资讯精选 75
Here are the top AI agents that can live in your text messages· 可集成到短信中的顶级 AI 助手
We created a list of the most notable AI agents that can live in your text messages, from …
行业资讯精选 75
西门子将亮相2026工博会：以软硬协同的工业全栈能力，推动物理AI加速落地
西门子将以更集中、更场景化的方式，展示其工业全栈能力
行业资讯精选 83
Nano Banana终于2.1了！4K直出，中文进步太明显
Nano Banana 2.1 更新，提升中文处理能力
行业资讯精选 85
JEPA、空间智能路线正面交锋：WorldArena 2.0 挑战赛开考「能不能被机器人真正用起来」| IROS 2026
<img data-src="https://static.leiphone.com/uploads/new/images/20261010/6ac9b2326d2ea.jpg" …
行业资讯精选 65
这届 IROS 2026 上，具身学术与产业边界正在消融
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/s…
行业资讯精选 85
阶跃 Step 5 Preview 登上 OpenRouter 全球第一，10月15日开源！
10 月 10 日消息，阶跃星辰最新旗舰基座模型 Step 5 Preview 登顶 OpenRouter Trending 榜，并位列日榜全球第二，成为近期全球开发者社区最受关注…
行业资讯精选 75
数据越多≠决策越好：从无人机滑翔到多机器人通信，重新审视数据与信息的关系 | IROS 2026
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/m…
行业资讯精选 82
让触觉像视觉一样可规模化，哥大李昀烛团队 FlexiTac 的开源实践 | IROS 2026
<img data-src="https://static.leiphone.com/uploads/new/images/20261010/6ac9afb68686f.jpg" …
行业资讯精选 70
2026中国国际社会公共安全产品博览会暨第三届智能与安全产业发展大会将于11月3日在北京开幕
中国国际社会公共安全产品博览会（以下简称“安博会”）是经中华人民共和国商务部批准，由中国安全防范产品行业协会主办的专业性展会。1994 年创办至今，依托行业主管部门、各地相关单位、…
行业资讯精选 85
GMIF 2026：日均140万亿Token调用，正在改变SSD的能力与分工
作者 | 陈悦琳 编辑 | 包永刚 <p style="text-align: left; margin-top: 6pt; margin-botto
行业资讯精选 75
Anthropic can’t reliably control its AI agents. It’s cutting off its internal evals from the live internet instead· Anthropic暂停AI实时互联网访问
Anthropic said it "turned off live internet access" for "all our internal evaluations" unt…
行业资讯精选 65
The maker of non-text AI model Jev valued at $7.5B just weeks after launch· 非文本AI模型Jev的制造商在推出数周后估值75亿美元
TypeSafe AI raised $870 million in a round led by a16Z.
行业资讯精选 85
An Anthropic AI model sent a false homicide tip to Philadelphia police· Anthropic AI 模型向费城警方提供错误杀人线索
Anthropic did not discover this behavior until over two months after its AI submitted the …
行业资讯精选 75
Amazon and others are done keeping data center deals secret. Is it enough to build trust?· 亚马逊等停止对数据中心交易保密
Amazon says it will stop using NDAs when negotiating data center deals with local governme…
行业资讯精选 81
Amazon drops data center NDAs, and AI agents want your credit card· 亚马逊放弃数据中心保密协议
Amazon says it will stop using NDAs when negotiating data center deals with local governme…
行业资讯精选 82
We can’t help treating AI like it’s human. But should we?· 我们为何将 AI 视为人类？
"When we are drawn into even the most primitive exchanges with a relational artifact, we b…
行业资讯精选 82
A16z’s Olivia Moore on the state of consumer AI· A16z 合伙人谈消费级 AI 现状
Moore sees a huge opportunity in consumer AI, particularly if the industry can tap into re…
行业资讯精选 83
当年「字节投毒实习生」田柯宇，估值 2 亿美元，要挑战李飞飞做世界模型
<section data-mpa-powered-by="yiban.io" data-pm-slice="0 0 []" style="-webkit-tap-highligh…
行业资讯精选 85
Asana cuts model costs 76x in browser tests with GPT-6.1 Sol· Asana 优化浏览器代理成本和速度
Using GPT-6 Astra in Codex, Asana made its browser agent 76x cheaper and 5x faster in test…
YouTube 热门 AI 视频
AI 视频精选 80
Supporting Clinicians With AI | Wenyi Zhu, Founder of Kuddo Health· AI 辅助临床医生
YouTube · NVIDIA
📺 NVIDIA▶ 91 次观看
AI 视频精选 80
Order in any language with Gemini 3.5 Live Translate· Gemini 3.5 实时翻译
YouTube · Google for Developers
📺 Google for Developers▶ 14 次观看
AI 视频精选 65
Set up your dot in the ChatGPT mobile app· 如何设置 ChatGPT 移动应用个人资料
YouTube · OpenAI
📺 OpenAI▶ 4.6k 次观看
AI 视频精选 85
🤖 Operating real hardware with Gemini AI· 用 Gemini AI 操作真实硬件
YouTube · Google for Developers
📺 Google for Developers▶ 155 次观看
AI 视频精选 75
Building the Codex community· 构建 Codex 社区
YouTube · OpenAI
📺 OpenAI▶ 557 次观看
AI 视频精选 82
Building low-latency remote robotics with Gemini· 使用 Gemini 构建低延迟远程机器人
YouTube · Google for Developers
📺 Google for Developers▶ 183 次观看