AI 情报文章归档

Reinforcement learning enables mobile robots to make fast and scalable navigation decisions in constrained human-shared environments. However, directly extending single-robot approaches to multi-robot settings often fails to ensure safety …

In this paper, we systematically investigate the mechanisms, scaling behavior, and generalization of physical Recursive self-improvement. We introduce PhysRSI-eval to quantify physical RSI through a five-axis metric: First-Attempt Performa…

Long-horizon agents must learn heterogeneous capabilities from interleaved trajectories. Updating shared parameters to learn new capabilities can degrade previously acquired ones, resulting in catastrophic forgetting. As a trajectory unfol…

A seemingly reasonable push in Sokoban can make the puzzle unsolvable. To avoid such mistakes, an agent has to anticipate the consequences of its actions before execution, especially when feedback from the real environment is sparse or act…

The goal of a world model is to capture the dynamics of an environment. This has the great appeal that it allows agents to plan and simulate a course of actions with the help of a world model. For these uses to be reliable, however, the le…

Systems that combine several large language models (LLMs) often use one context for many questions. When the context is passed between models as text, the receiving model processes all of its tokens again for every question. Avoiding this …

Bilevel optimization provides a framework for machine learning tasks such as hyperparameter optimization, data reweighting, and meta-learning. Existing Newton-based bilevel methods typically rely on the classical hypergradient formula, whi…

LLM agents that compute the right answer can still answer wrongly. Agent harnesses increasingly keep long tool outputs out of the context window, saving them to files that the agent searches or reads on demand. We ask where this costs accu…

We propose KronLSTM, a linear attention architecture based on the Kronecker Delta Rule, a novel update rule for associative matrix memories. KronLSTM extends the xLSTM family with a memory-revision mechanism that corrects targeted associat…

Machine learning is widely regarded as a promising application in which quantum computing could offer a computational advantage. However, general quantum machine learning models with provable guarantees on performance, trainability, and ef…

I built Sorcerer, a free, open-source macOS app (SwiftUI, macOS 15+, MIT) that simulates a community...

Many modern language models have adopted hybrid architectures that incorporate both linear-time recurrence and softmax attention. Modern recurrence layers provide a highly expressive mixing mechanism while ensuring high throughput by compr…

Frontier agents retrieve structural metadata well and compute geometry from coordinates less reliably. We ask how much of a benchmark score reflects that gap and how much reflects the evaluation configuration around the model. We study Pro…

Agent learning is shifting from learning from human data toward learning through interaction with the world. However, scaling such learning directly in real-world computer systems is challenging and can introduce substantial security risks…

Context compression improves long-context efficiency but may introduce faithfulness issues through an information bottleneck. Existing evaluations rely on QA metrics, but a reader can still answer correctly from an unfaithful compression, …

Beyond basic tool use and coding skills, effectively incorporating visual references into agentic generation is equally essential for applying AI agents to real-world productivity tasks. We study this crucial capability through webpage dev…

Memory-native evolution requires making memory an intrinsic capability of foundation models, enabling them to continuously accumulate and utilize information. To this end, we introduce MemT-1.0, a new architecture of memory foundation mode…

恭喜 @jang_yoel 和团队！ 做梦的好时代啊 今天，Dream Labs 正式走出隐身模式，推出 Dream-1——一个机器人基础模型，能够控制它从未见过的双臂机器人，完成从未被训练过的任务，在从未到过的场所，每小时仅需 3 美元。 Dream API 即日起对所有人开放。🧵

Exa 发布 Exa 3.0，其 Search API 新增 Dynamic Highlights 模式，在共享预算内决定每页返回的文本量。

BBC 报道，美国检察官称涉嫌在 OpenAI 高管会议室安装窃听器的前工程师 Alfozan 获得沙特政府支持。一名未具名沙特王室成员与其频繁 WhatsApp 联系并试图招募其加入沙特 AI 项目，Alfozan 曾在 5 月同意“沙特比美国更需要他”，同时表示要等 OpenAI 股票归属；被捕次日其妻儿乘沙特军官预订的商务舱单程飞往卡塔尔。

Kling AI 发布视频生成模型 Kling 4.0，单次可生成最长 30 秒视频，提示词理解扩展到 8000 token，Omni Reference 最多接受 15 个参考素材，新增双声道立体声和十余种语言的对口型。

蚂蚁 inclusionAI 在 Hugging Face 和 ModelScope 发布医疗专用模型 Ling-3.0-flash-Sante，基于 Ling-3.0-flash 构建，总参数 124B、每 token 激活 5.1B。

“对大多数AI算力提供商来说太低了，无法收回其AI投资” Gartner：许多建设AI数据中心的企业，可能在芯片过时之前都赚不回投入的成本。 转售正成为GPU经济的一部分，@shadeformai Spot让供应商能把闲置算力重新投放市场。 Shadeform的Flex VM可以接收来自任何数据中心的裸金属GPU，并在30分钟内使其可租用。 小型运营商从第一天起，就和其他人拥有同样的入口。

OpenAI 首席科学家 Jakub Pachocki 宣布 Aleksander Madry 回归 OpenAI，与他就 AI 进展的速度与影响及对公司乃至全行业战略的启示展开工作。Madry 此前离职几个月后表示这更像一次学术休假，认为在 AI 进展进入新阶段后，应对越来越强的 AI 带来的安全与安保挑战必须成为核心优先事项，他将加入 Pachocki 团队协助规划应对路径。

@reflection_ai @baseten - 排行榜：https://arcprize.org/leaderboard - 复现公开结果：https://github.com/arcprize/arc-agi-benchmarking - 测试政策：https://arcprize.org/policy - Beam 1.0 完整结果：https://arcprize.org/results/reflection-beam-1-0

@reflection_ai 每任务成本基于我们测试时使用的 @Baseten 定价：每 100 万 token 输入 $1.30 / 输出 $4.30。 ARC-AGI-3 结果待公布。 完整结果：https://arcprize.org/results/reflection-beam-1-0

ARC Prize 公布 Beam 1.0 在 ARC-AGI (Verified) 的成绩：ARC-AGI-2 为 23.5%，每任务 $0.61；ARC-AGI-1 为 74.0%，每任务 $0.31。最佳成绩出现在 xhigh 推理档，但 high 档在 ARC-AGI-2 上也达到 20.0%，成本约低 43%。

你一直在用 Codex 做什么？

Salesforce 论文研究跨委托方的智能体间（A2A）协作失败原因，基于 Claude Code、Codex 和 Cursor 在独立沙箱中跑了 2500 多场会话、488 个场景，归纳出 20 种失败模式。

如果我在斯洛文尼亚办一场聚会，把它办得超炸，会怎样？

Ars Technica 报道称，AI 生成的游戏克隆让多位独立开发者抱怨作品被整批复制，McPixel 开发者 Sos Sosowski 记录了诸多案例，指出开发者陷入不分享就没销量、分享就被抄袭的两难。

Inception 的扩散模型 Mercury 3 上线 OpenRouter，生成速度超 1,000 tok/s，支持 1M 上下文，是 Mercury 2.5 的 4 倍。Inception 报告其在 IFBench、GPQA Diamond 和 SWE-Bench Atlas 上超过 GPT-6 Luna 和 Haiku 5.5；上线期五折，价格 $0.10/M 输入、$0.35/M 输出。

如果你的 AI 预算流向最会用它的工程师，会怎样？ 一刀切的使用上限，是变相的招聘冻结。 为最高产的工程师推出基于效率的自动审批，外加 Devin Coach 帮助每个人达到这一水平。 https://devin.ai/blog/efficiency-based-cost-controls

你手上的事，我都帮你接过来

面向打包多模态掩码的精确序列并行。 它按层选择 ring attention、稀疏 all-to-all 或 Ulysses，数值与单 GPU 一致，并在最多 128 块 H100 上于 48 个案例中的 47 个优于 FlashMask + Ulysses 混合方案。 https://lumalabs.ai/news/packed-multimodal-attention

Wabi 面向公众上线其 AI messenger 应用，用户可将 AI 助手加入与亲友的群聊。助手可监听群聊内容并按需行动，例如日程变更时更新共享日历；计划可拆分为 threads，能创建自定义工具如餐厅地图；可连接 Gmail、Google Calendar 等账户并主动发现问题和发起对话，重要操作需用户许可，用户可随时移除连接。

没有身体的大脑只会空谈。 有身体的大脑才能把事情做成。 LLM（模型）是大脑。harness 是身体：围绕模型的工具、文件和循环。Agent SDK 让你为自己的用例定义身体。

DeepSeek V4.1 Flash 在 OpenCode Go 和 Free 中已更新至 1017 你这边无需做任何改动

AgentCloak 创始人 Peter Yared 在镜头前砸碎一台 Mac Mini，这条发布视频获得 220 万次播放和 3000 次下载。launchvideo.com 以流水线方式操盘此次发布：创始人出镜视频，创作者按受众规模分批放出。团队还拍了备选结尾——配置 1 万美元 Mac Studio 后合上笔记本，但砸机版本保留为主镜头，备选未使用。

Grok @Bot 可能会成为你最喜欢的员工