AI 情报文章归档

Arena 宣布 Meta 的 Muse Spark Open 已进入 Agent Arena，用户可实测，评分即将公布。该模型为 816B 总参数、42B 激活参数，1M token 上下文窗口，原生多模态。

微软 MAI 宣布 MAI-Base-2 预训练完成，MAI-Base-3 训练周期已在筹备中。MAI-Base-2 在留出评测上超越所有受测开放权重模型，预训练 FLOPs 效率比最强开放基座模型 DeepSeek-V4-Pro-Base 高 4-12 倍，节省的算力被用于训练更大模型。微软称已在 MAI 内部建成 3T 参数级别的 SOTA 预训练基座模型，后训练结果即将公布。

Meta 发布 Muse Spark Open AI 模型，称其为迄今最强的美国开放权重模型，权重今日放出。Meta 称该模型是同类中首个可在桌面端用两台 Apple M5 Ultra Mac Studio 配合最新 MLX 框架运行的模型。

Thomas Wolf 转发并评论 Muse Spark Open 的发布，称其为迄今最强美国开源权重模型。引用内容称该模型总参数 816B、激活 42B、上下文窗口 1M token、原生多模态，面向编码与长程智能体任务；配图显示其在 JobBench 61.8、HLE-Diamond 26.1、Terminal-Bench 2.1 76.4 等基准上领先所列对比模型。

开始探索吧。 https://fal.ai/worlds

fal 宣布 fal Worlds 由其专为世界模型构建的推理引擎 fal World Model Accelerator 驱动。该引擎提供低延迟推理、实时流式和可扩展的基础设施，开发者可以借此让交互式世界落地，用户可实时探索、采取行动并改变环境。详情见 https://fal.ai/docs/documentation/development/wma

fal 推出 fal Worlds，一种实时创建并交互 AI 生成世界的新方式。用户可构建角色、探索环境、导演场景，世界会随用户行动实时响应，并支持上手体验 LingBot World、H3 Max Director 等最新前沿世界模型。

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle ...

Bland 发布 Blandie 公开演示，可根据企业网站自动生成带品牌名称和 logo 的语音与短信智能体，能处理预订和支付。消息以企业品牌身份通过 Android 的 RCS 和 iPhone 的 iMessage 发送，回答基于网站内容，还能回复网站截图和图片，并记住客户是否拨打过 Bland 电话。

各位，这就像我的超级碗，我等了太久，美国开源权重模型终于再次有竞争力了。在今天之前，美国模型仍然落后于最早2026年2月的中国模型。还有更多工作要做！ 这把差距至少缩小了一半，按时间算（大约3个月）

NVIDIA 的 Nemotron 3.5 Super VL 现已在 Perplexity 上面向 Pro 和 Max 订阅用户开放，并已接入 Agent API。

Meta 发布开放权重模型 Muse Spark Open，现已可在 OpenRouter 使用，定位为面向编码与智能体工作的美国最强开放权重模型。模型总参数 816B、激活 42B，支持 1M token 上下文窗口，原生多模态，OpenRouter 定价为输入 $1.25/M、输出 $4.25/M，正文提到支持图像、视频、音频和 PDF 输入。

Self Labs 发布 Agent Vault，一个托管浏览器代理，让 Claude Code、Codex 和 Grok Bot 等智能体用占位符令牌完成登录和信用卡支付，代理只在填写瞬间向网站输入真实卡号或密码。由于智能体本身不持有可窃取的凭证，被提示词注入的智能体最多只能在用户设定的额度内消费。作者评价这是智能体处理支付应有的方式。

GitHub 用原生功能构建的 GitHub App Evergreen，通过两个工作流自动化组织卫生：每季度审查外部协作者访问权限，未响应即默认撤销；检测并归档长期无人类活动的仓库，且从不删除仓库内容。它由 Go webhook 服务器驱动，状态以 JSON 文件提交到 controller repo，用 git 历史充当审计记录，决策通过仓库内的 checklist issue 完成。

TechCrunch Founder Summit 的 Founder 与 Investor Pass 早鸟优惠将于 10 月 16 日结束，最高可省 200 美元，Founder Pass 现价 199 美元、Investor Pass 现价 399 美元。

Personalization is essential for general-purpose robots to transition into household environments and achieve true practicality. While general robotic manipulation has been extensively studied, personalized robotic manipulation (PeRM) lack…

Significant efforts are currently being directed toward the implementation of large-scale quantum information processors, but realizing useful quantum advantage also requires the development of efficient quantum algorithms. While for many …

Predicting how the world responds to an action requires more than what can be seen: robots must account for environmental factors that are neither directly observable nor explicitly specified, from physical properties such as friction and …

Generative simulation can automate robot environment construction, but physically valid environments are not necessarily useful for training. The challenge is to identify environments that address a policy's failures and supervision that h…

Time series foundation models are becoming increasingly popular for general-purpose forecasting. However, most leading models predict marginal quantiles, leaving the dependence structure of future trajectories unspecified. This distinction…

Generative visual verifiers play a crucial role in advancing multimodal intelligence. To enhance the robustness of visual verifiers in complex multimodal scenarios involving compositional requirements, we move beyond conventional binary ve…

Mixture-of-experts (MoE) models have become an important approach for scaling model parameters. Modern MoE architectures increasingly combine two design choices: fine-grained expert segmentation and shared experts. The former increases exp…

Black Forest Labs 在 ComfyUI 发布 FLUX 3 家族的紧凑开源权重模型 FLUX 3 Image Klein，共 5.6B 参数，提供 Base 和 Distilled 两种版本，内部评测中文生图和多参考生成得分高于 Qwen-Image 2.1。

Existing agent benchmarks typically evaluate agents on well-specified tasks defined by issue reports, specifications, or task tickets. Real-world customer delivery is different: forward-deployed engineers (FDEs) must clarify ambiguous requ…

Large language models (LLMs) are expanding from assisting scientific research and engineering to autonomously developing and improving AI systems, advancing AI for AI (AI4AI) as a pathway toward recursive self-improvement (RSI). Workflow-d…

Recent 3D world models increasingly use features from geometry foundation models (GFMs) as predictive world states, benefiting from their rich geometric priors. However, the view dependence of these features entangles scene dynamics with c…

We introduce 3DCodeVerse, a research ecosystem that enables coding agents to create editable and verifiable 3D content at scale. 3DCodeVerse comprises large-scale, high-quality datasets of executable 3D code, a coding harness for spatial r…

Meta 发布开源权重模型 Muse Spark Open，已在 Hugging Face 以 Muse Model License 提供。模型总参数 816B、激活 42B，上下文窗口 1M token，面向编码与智能体工作流，原生多模态（图像、音频、文档），并相比 8 月发布的 Muse Spark 增加了安全后训练。

How can robots autonomously collect the contact-rich experience needed to learn dexterous manipulation? We present DSP: Dexterity from supervised play, a framework in which a robot learns a play policy from minutes of human play through be…

Boris Cherny 转发 Gray Swan 与 Anthropic、Meta、Google DeepMind、Reflection 等合作举办的间接提示词注入红队挑战赛，总奖金池 30 万美元。比赛名为 Indirect Prompt Injection: Black，分三波于 10 月 15 日、10 月 22 日和 10 月 29 日进行。Cherny 表示 Claude 模型在提示词注入方面近年已有显著进展，随着智能体承担更多任务，全行业解决该问题愈发紧迫。

Retrieving relevant lemmas is crucial for formal theorem proving, enabling provers to build machine-checkable proofs from established mathematical results. Recent retrieval systems match library declarations to a target theorem or its inte…

Hugging Face CEO Clément Delangue 演示通过命令 llama serve -hf ggml-org/Muse-Spark-Open-816B-A42B-GGUF:Q2_K，在两台 Mac Studio 上运行该 816B-A42B GGUF 量化模型，并向 finkd 与 johnternus 致意。

了解更多 ↓ https://goo.gle/4jfSznG

Gemini 4 Argon 能在几分钟内把杂乱数据变成干净的可视化叙事 🪐 它处理了 @NASA 开放档案中的 80,000 个文件（CSV、图像、PDF、网页文章、XML 日志和原始视频），并依据严格的设计规范生成了一份 Google Slides 演示文稿。

We develop a robust reinforcement learning (RL) framework for learning adaptive treatment policies that generalize across patients and remain reliable under errors in the measured biomarker and shifts in the disease dynamics. By modeling t…

Large language models (LLMs) can often tell when they are being tested. This ability, known as evaluation awareness, can lead models to act more cautiously during safety evaluations than in real use. Existing work has studied evaluation aw…

很喜欢 Meta 开源了 Muse Spark Open！ 大概是 Muse Spark 1.2 的水平。 我最大的愿望：Meta 后续能推出 Muse Spark 1.3 的开源权重。 那将是世界上最好的开源模型。

Deep search agents solve knowledge-intensive tasks through long-horizon, multi-turn interactions, where assigning credit to individual search decisions is fundamentally difficult. Outcome-based reinforcement learning provides only a single…

Continuous diffusion language models (CDLMs) offer an alternative approach to parallel language generation; however, most existing studies directly use transformer architectures from either causal large language models (LLMs) or Diffusion …

Image editing models produce impressive results for individual edits, but their quality often deteriorates over multiple editing turns. Each editing turn introduces small distribution shifts that accumulate over successive edits, creating …