AI 论文与研究

Magic-W0 是一个联合建模结构化物理状态演化与连续动作的世界-动作基础模型，将交互表示为包含 Current State、Transition 和 Future State 的 Structured World Transition。

如果这个月只能读一篇文章，就是这篇！ 如果自动化 AI 研发触发了智能爆炸，会怎样？ @geoffreyhinton： 由递归自我改进引发智能爆炸的想法已经存在很久，但直到最近它似乎都不迫在眉睫。如今许多顶尖研究者认为它可能很快就会发生。你可以在这里读到我们关于此的论文： https://casp.ac/reports/intelligence-explosion

国际清算银行（BIS）发布关于 AI 公司间循环融资的报告：2021 至 2025 年间，AI 公司获得的投资中 55.2% 来自其他 AI 公司，同时 AI 投资者将自身 28.7% 的交易额投向 AI 标的。循环交易仅占 AI-to-AI 交易的 16.1%，却持有其中 46.4% 的资金；芯片、云和基础设施供应商参与了 73% 的循环关系。报告提醒，供应商投资客户可能同时损失股权和未来订单，且风险分散在多国多行业，没有单一监管者能看到全貌。作者补充称美国 AI 巨头的…

在我们的研究博客上阅读更多内容：https://research.meta.ai/blog/solving-open-research-problems-together

6️⃣ 非结合代数：团队与 Muse Spark 合作，为一条受生物学启发提出的数学结构规则找到了一个例外。他们更进一步，发展出一种替代性刻画，并由研究人员进行了验证和完善。 阅读论文：https://ai.meta.com/research/publications/on-solvable-evolution-algebras-and-a-conjecture-by-garcia-martinez-and-perez-rodriguez/

4️⃣ 优化：什么时候可以用更简单的问题替换一个困难的数学问题，而不丢失任何东西？研究人员与 Muse Spark 合作，证明了一条明确的规则，说明某种特定简化何时能精确捕捉原问题，何时会留下缺口。 阅读论文：https://ai.meta.com/research/publications/tightness-of-the-cycle-based-relaxation-for-completed-length-three-alpha-cycles/

3️⃣ 群论：研究人员通过找到一个反例，推翻了一条关于描述对称性的数学结构的猜想。Muse Spark 生成了找到该反例的搜索代码，团队验证了结果并完成了证明。 阅读论文：https://ai.meta.com/research/publications/semiabelian-groups-need-not-be-monomial/

1️⃣ 概率：数学家与 Muse Spark 合作，回答了关于将随机点拟合到拉伸球面上的问题。在所研究的设定下，他们证明了精确拟合可能与否之间存在一个明确的临界点。 阅读论文：https://ai.meta.com/research/publications/the-strict-threshold-for-gaussian-ellipsoid-fitting/

2️⃣ 微分方程：想象一场拔河——一种效应把波向内挤压，另一种效应把波向外扩散。波能永远聚集下去吗？在 Muse Spark 的帮助下，研究人员证明，在所研究的条件下，激光启发模型中的波必须在有限时间内"爆破"。 阅读论文：https://ai.meta.com/research/publications/finite-time-blow-up-of-radial-negative-energy-solutions-for-the-mass-critical-biharmo…

Meta 分享数学家与 Muse Spark 1.1 和 Muse Spark 1.2（Thinking Mode）在 meta.ai 普通聊天界面下协作完成的六篇论文，面向无现成解法的开放数学问题，未使用定制研究脚手架。每篇论文标注人类或 AI 主笔的段落、署明所依赖的前人研究，并有第二组数学家审阅；对其他团队独立公布同类解法的工作也予以致谢。

Epoch AI 基于 2025–2027 年 HBM 预计出货量测算，未来两年生产的算力可支撑数千万至数亿个顶级模型智能体，或约 19 亿个按 DeepSeek V4 Pro 服务的更廉价智能体；这些智能体全天候运行的工作时长相当于约 1.4 亿至 7 亿名全职员工。

Epoch AI 发布估算智能体人口规模的报告及代码，提出 AI 硬件建设实际上是对经济的押注，即人们会以大约相当于人类工资的小时成本，付费让 AI 在经济各领域执行有价值的任务。分析由 @genericname2134 完成，完整报告见 https://epoch.ai/publications/estimating-the-agent-population。

为估算智能体群体的规模，我们首先确定智能体连续工作一小时的成本，以便与人类工资相比较。 在我们分析的智能体会话中，按 API 价格计算，Codex 工作负载平均约为每智能体小时 $16–$18，而 Claude Code 约为 $24–$50。

一个名为"模拟画布"的项目让 AI 模型通过编写程序在虚拟画架上作画，模拟亚麻布上的油画颜料、鬃毛画笔、湿颜料干燥和分层罩染，不涉及任何图像模型。

Redwood Research 发文指出，将安全研究定义为"在不显著增加成本或降低有用性的前提下让 AI 部署更安全"，会把几乎所有能力研究都算作安全研究，因为任何研究都在扩展安全-有用性帕累托前沿。作者认为关键在于开发者必须在帕累托前沿上选点：安全研究通常引导开发者选择更高安全，能力研究则相反，但若未来政治意愿大幅提高，某些能力研究也可能成为提升安全的有效途径。

This paper examines the role of Novel View Synthesis (NVS) in geometric representation learning. In principle, NVS should reason about 3D scene structure, thereby enabling transferable multi-view geometric representations. Yet, existing en…

We introduce 4DCodeBench, a benchmark for 4D inverse graphics through code generation, in which agents reconstruct dynamic scenes from video as executable graphics programs. To accomplish this, agents must translate visual observations int…

Continual learning treats degradation on previously seen data as evidence of failure, a convention inherited from settings with a stationary prediction target, where a correct label remains correct indefinitely. World models do not satisfy…

Ribonucleic acid (RNA) functions through conformational changes that are not fully captured by static structures. However, large-scale standardized RNA dynamics data remain limited, and existing approaches typically treat trajectory genera…

Inspired by human vision, we introduce a framework using active gaze to enable fine-grained bimanual manipulation with only a single stereo camera. EyeRobot 2.0 physically attends to a 3D fixation point in the scene by swiveling two eye vi…

We study the minimization of sums of smooth strongly convex functions over undirected graphs, with each function held by one agent and communication restricted to neighbors in the graph. Existing decentralized methods, whether based on gos…

The choice of post-training data for large language models substantially affects downstream performance. Gradient-based data selection is a popular approach that ranks training data by how well their gradients align with those of a small v…

Modern chess engines are silent experts: they play at a superhuman level, but do not offer explanations for their play. On the other hand, language models (LMs) can generate plausible-sounding explanations, but their weak playing strength …

Adaptive Reward Routing 通过前向过程 RL 实现音视频联合扩散的动态多奖励优化 论文：https://huggingface.co/papers/2609.37200

Scarcity of labeled data limits development of deep learning biomarkers in oncology. We develop a two-stage AI model predicting pathological complete response (pCR) to neoadjuvant therapy in breast cancer. The first stage learns the transc…

The dynamics of cells, organisms, and fluids are often modeled as probability distributions evolving over time. Reconstructing and extrapolating this evolution from unpaired snapshots requires assumptions about the underlying process. Wass…

LLM-guided evolutionary methods, such as AlphaEvolve, have emerged as powerful approaches for challenging computational optimization problems, such as circle packing. However, prior work typically optimizes performance gain over a fixed nu…

Policy-gradient methods are central to modern reinforcement learning, including LLM post-training. When they struggle, the usual suspects are exploration, credit assignment and action-sampling noise. Classification has none of them. A clas…

Masked diffusion language models (dLMs) offer a promising parallel alternative to autoregressive models for complex reasoning. However, they face a distinct credit-assignment challenge, since a few commitments during denoising sharply redu…

Deploying a new decision policy creates a cold-start problem for prediction models whose targets depend on the policy's actions: historical observations reflect earlier policies, while real observations under the new policy are not yet ava…

Multi-pitch estimation in vocal ensembles is challenging because singers occupy overlapping pitch ranges and often sing at closely spaced fundamental frequencies, causing their harmonics to overlap in time-frequency representations. Existi…

Dense-retrieval services must switch among embedding-prefix dimensions and index bit rates as latency, quality, and memory budgets change. Tuning a quantizer separately for each rate gives the best quality, but the retrieval tier then hold…

Federated Learning (FL) is a privacy-oriented learning paradigm that enables collaborative model training while keeping training data local to participating clients. However, it does not guarantee that clients submit policy-compliant contr…

Marine ecosystems are impacted by various threats such as oil spills, algal blooms, and sediment floods, which disrupt habitats, wildlife, and human activities. Advances in satellite imagery and Artificial Intelligence (AI) have enhanced o…

Many machine learning methods aim to approximate the lower-dimensional manifold on which the data lives. A desirable feature of such methods is that they should capture the epistemic uncertainty of this learned manifold. One model that ach…

Stationarity rewards memory, but after a change the same history can mislead. We ask when forgetting should be permitted. E-process-authorized Thompson sampling (e-ATS) gives each arm full-history and discounted Beta states. An anytime-val…

EgoTools 面向真实世界自我中心视频中以工具为中心的推理 paper: https://huggingface.co/papers/2609.39378

Google Research 发布 Cogentic，一个基于 Gemini 的多智能体自动证明发现框架，面向理论计算机科学开放问题，只需问题陈述即可自动产出完整论文形式的结果。

了解更多 ↓ https://blog.google/innovation-and-ai/models-and-research/google-research/google-project-suncatcher-facts/?utm_source=tw&utm_medium=social&utm_campaign=og

NVIDIA 发布 Long-Transduction 测试框架，让模型在数千次输出中持续读取、更新并输出状态相关结果，分别变化三个因素。在七个开源权重模型上，上下文从 4K 增至 128K 时准确率下降 62.8%，仅改变输入格式下降 36.5%，单步操作变难下降 39.9%。