OrangeBot.AI Digest — 2025-12-27

All Algorithms implemented in Python

结合LightRAG 知识库的知识图谱智能体平台。 An agent platform that integrates a LightRAG knowledge base and knowledge graphs. Build with LangChain v1 + Vue + FastAPI, support DeepAgents、MinerU PDF、Neo4j 、MCP.

Expose your local web server to the internet with a public URL.

Collection of awesome LLM apps with AI Agents and RAG using OpenAI, Anthropic, Gemini and opensource models.

CV/resume generator for academics and engineers, YAML to PDF

While Large Multimodal Models (LMMs) have made significant progress, they remain largely text-centric, relying on language as their core reasoning modality. As a result, they are limited in their ability to handle reasoning tasks that are predominantly visual. Recent approaches have sought to address this by supervising intermediate visual steps with helper images, depth maps, or image crops. However, these strategies impose restrictive priors on what "useful" visual abstractions look like, add heavy annotation costs, and struggle to generalize across tasks. To address this critical limitation, we propose a task-agnostic mechanism that trains LMMs to discover and use visual reasoning tokens without explicit supervision. These tokens attend globally and re-encode the image in a task-adaptive way, enabling the model to extract relevant visual information without hand-crafted supervision. Our approach outperforms direct fine-tuning and achieves state-of-the-art results on a diverse range of vision-centric tasks -- including those where intermediate abstractions are hard to specify -- while also generalizing to multi-task instruction tuning.

Large-scale autoregressive models pretrained on next-token prediction and finetuned with reinforcement learning (RL) have achieved unprecedented success on many problem domains. During RL, these models explore by generating new outputs, one token at a time. However, sampling actions token-by-token can result in highly inefficient learning, particularly when rewards are sparse. Here, we show that it is possible to overcome this problem by acting and exploring within the internal representations of an autoregressive model. Specifically, to discover temporally-abstract actions, we introduce a higher-order, non-causal sequence model whose outputs control the residual stream activations of a base autoregressive model. On grid world and MuJoCo-based tasks with hierarchical structure, we find that the higher-order model learns to compress long activation sequence chunks onto internal controllers. Critically, each controller executes a sequence of behaviorally meaningful actions that unfold over long timescales and are accompanied with a learned termination condition, such that composing multiple controllers over time leads to efficient exploration on novel tasks. We show that direct internal controller reinforcement, a process we term "internal RL", enables learning from sparse rewards in cases where standard RL finetuning fails. Our results demonstrate the benefits of latent action generation and reinforcement in autoregressive models, suggesting internal RL as a promising avenue for realizing hierarchical RL within foundation models.

Existing video generation models struggle to maintain long-term spatial and temporal consistency due to the dense, high-dimensional nature of video signals. To overcome this limitation, we propose Spatia, a spatial memory-aware video generation framework that explicitly preserves a 3D scene point cloud as persistent spatial memory. Spatia iteratively generates video clips conditioned on this spatial memory and continuously updates it through visual SLAM. This dynamic-static disentanglement design enhances spatial consistency throughout the generation process while preserving the model's ability to produce realistic dynamic entities. Furthermore, Spatia enables applications such as explicit camera control and 3D-aware interactive editing, providing a geometrically grounded framework for scalable, memory-driven video generation.

Large language models increasingly expose reasoning traces, yet their underlying cognitive structure and steps remain difficult to identify and analyze beyond surface-level statistics. We adopt Schoenfeld's Episode Theory as an inductive, intermediate-scale lens and introduce ThinkARM (Anatomy of Reasoning in Models), a scalable framework that explicitly abstracts reasoning traces into functional reasoning steps such as Analysis, Explore, Implement, Verify, etc. When applied to mathematical problem solving by diverse models, this abstraction reveals reproducible thinking dynamics and structural differences between reasoning and non-reasoning models, which are not apparent from token-level views. We further present two diagnostic case studies showing that exploration functions as a critical branching step associated with correctness, and that efficiency-oriented methods selectively suppress evaluative feedback steps rather than uniformly shortening responses. Together, our results demonstrate that episode-level representations make reasoning steps explicit, enabling systematic analysis of how reasoning is structured, stabilized, and altered in modern language models.

Videos are continuous 2D projections of 3D worlds. After training on large video data, will global 3D understanding naturally emerge? We study this by quantifying the 3D understanding of existing Video Foundation Models (VidFMs) pretrained on vast video data. We propose the first model-agnostic framework that measures the 3D awareness of various VidFMs by estimating multiple 3D properties from their features via shallow read-outs. Our study presents meaningful findings regarding the 3D awareness of VidFMs on multiple axes. In particular, we show that state-of-the-art video generation models exhibit a strong understanding of 3D objects and scenes, despite not being trained on any 3D data. Such understanding can even surpass that of large expert models specifically trained for 3D tasks. Our findings, together with the 3D benchmarking of major VidFMs, provide valuable observations for building scalable 3D models.

Autoregressive (AR) visual generation relies on tokenizers to map images to and from discrete sequences. However, tokenizers are trained to reconstruct clean images from ground-truth tokens, while AR generators are optimized only for token likelihood. This misalignment leads to generated token sequences that may decode into low-quality images, without direct supervision from the pixel space. We propose VA-π, a lightweight post-training framework that directly optimizes AR models with a principled pixel-space objective. VA-π formulates the generator-tokenizer alignment as a variational optimization, deriving an evidence lower bound (ELBO) that unifies pixel reconstruction and autoregressive modeling. To optimize under the discrete token space, VA-π introduces a reinforcement-based alignment strategy that treats the AR generator as a policy, uses pixel-space reconstruction quality as its intrinsic reward. The reward is measured by how well the predicted token sequences can reconstruct the original image under teacher forcing, giving the model direct pixel-level guidance without expensive free-running sampling. The regularization term of the ELBO serves as a natural regularizer, maintaining distributional consistency of tokens. VA-π enables rapid adaptation of existing AR generators, without neither tokenizer retraining nor external reward models. With only 1% ImageNet-1K data and 25 minutes of tuning, it reduces FID from 14.36 to 7.65 and improves IS from 86.55 to 116.70 on LlamaGen-XXL, while also yielding notable gains in the text-to-image task on GenEval for both visual generation model (LlamaGen: from 0.306 to 0.339) and unified multi-modal model (Janus-Pro: from 0.725 to 0.744). Code is available at https://github.com/Lil-Shake/VA-Pi.

Multi-turn reinforcement learning (RL) for multi-modal agents built upon vision-language models (VLMs) is hampered by sparse rewards and long-horizon credit assignment. Recent methods densify the reward by querying a teacher that provides step-level feedback, e.g., Guided Thought Reinforcement (GTR) and On-Policy Distillation, but rely on costly, often privileged models as the teacher, limiting practicality and reproducibility. We introduce GTR-Turbo, a highly efficient upgrade to GTR, which matches the performance without training or querying an expensive teacher model. Specifically, GTR-Turbo merges the weights of checkpoints produced during the ongoing RL training, and then uses this merged model as a "free" teacher to guide the subsequent RL via supervised fine-tuning or soft logit distillation. This design removes dependence on privileged VLMs (e.g., GPT or Gemini), mitigates the "entropy collapse" observed in prior work, and keeps training stable. Across diverse visual agentic tasks, GTR-Turbo improves the accuracy of the baseline model by 10-30% while reducing wall-clock training time by 50% and compute cost by 60% relative to GTR.

MIT 与丹娜法伯癌症研究院的科学家合作，首次在实验室成功合成了天然真菌分子“轮枝孢菌素A（verticillin A）”。该分子 50 多年前被首次发现，因其显著的抗癌潜力备受关注，但复杂的结构令其人工合成一直未能实现。研究成果发表于《美国化学会志》，有望开辟一类全新的抗癌药物研发路径。在最新研究中，研究团队不仅实现了轮枝孢菌素A的全合成，还以此为基础设计出多种新型衍生物。初步测试显示，部分衍生物对一种罕见的儿童脑癌——弥漫性中线神经胶质瘤表现出强大的抗肿瘤活性。研究团队从氨基酸衍生物β-羟色氨酸出发，逐步引入醇、酮、酰胺等化学官能团，并精准控制每一步的立体构型。历经16步精密反应，他们最终构建出轮枝孢菌素A分子。

国防科技大学透露，该校磁悬浮团队成功在两秒内，将吨级重的试验车加速至 700 公里/小时，测试速度打破了同类型平台全球记录，成为全球最快的超导电动磁悬浮试验速度。磁悬浮列车速度更快、加速减速更出色，维护成本更低，但建造成本更高，而且不兼容现有铁路设施。至今只有七列磁悬浮列车在运行——中国四列，韩国两列，日本一列。有两条城际磁悬浮线路正在建造，其中一条连接日本东京和名古屋，另一条连接湖南长沙和浏阳。日本实验性磁悬浮列车 L0 Series 曾在 2015 年创造了 603 km/h 的速度记录。

流行 GLP-1 减肥药如 Ozempic 不仅能帮助我们减轻体重，还会悄悄重塑我们的购物习惯，减少购买食品。根据发表在《Journal of Marketing Research》期刊上的研究，研究人员发现，服用 GLP-1 的家庭在六个月内食品杂货支出减少 5.3%。高收入家庭降幅更大为 8.2%。其中咸味零食的支出降幅最大。服用 GLP-1 的家庭在食品上的支出平均减少 10.1%。值得一提的是 GLP-1 服用者在酸奶和新鲜水果等健康食品上增加了支出。然而如果停止服用 GLP-1，研究人员观察到他们很快又恢复了过去的购买习惯，他们会在几个月内恢复大部分减掉的体重。

比特币挖矿难度在 2024 年翻倍，它的币值从今年 10 月创下的 12 万美元峰值跌至不到 9 万美元。尽管如此，比特币矿场的 ETF 今年飙升了约 90%，原因不是比特币，而是因为矿场纷纷转型 AI 数据中心。AI 竞争所亟需的资产恰好比特币矿场都有：数据中心、冷却系统、土地以及电力合同。当然 AI 数据中心需要更先进的冷却和网络系统，需要用英伟达的 GPU 替换专用矿机，但通过与矿场合作，AI 公司利用现有设施比从零开始建造新数据中心更快更便宜。以 Core Scientific 矿场为例，该公司认为转型为 AI 数据中心是难以想象的极佳机遇，它计划 2028 年完全退出比特币挖矿业务。

现代宇宙学的基础建立在宇宙学原理的假设之上，认为宇宙在大尺度是均匀对称的，在任何地方、任何方向上看起来都一样。然而一项强而有力的最新证据显示，这个基本假设可能是错误的。一个被称为宇宙偶极异常（Cosmic Dipole Anomaly）的谜团正挑战对宇宙的理解。过去科学家早已观测到宇宙微波背景辐射（CMB）存在明确的偶极现象，天空的一侧温度略高，而另一侧则略低，差异约为千分之一。这个现象被普遍认为是运动学效应导致，也就是太阳系、银河系乃至整个本星系群，正以每秒数百公里的速度在宇宙中穿梭。1984 年天文学家 George Ellis 和 John Baldwin 提出了一项检验方法。他们指出，若我们的运动是造成 CMB 偶极的唯一原因，那么这种运动也应该在遥远天体的空间分布上留下一个相对应的偶极，这项检验方法被称为 Ellis & Baldwin test。由于相对论效应，科学家预期在物质分布上观测到的偶极讯号其幅度不仅是与 CMB 偶极相同，还会根据天体数量、光谱特性相关的因素放大。然而异常之处就在此：尽管观测到物质分布的偶极方向与 CMB 偶极的方向一致，但其幅度却异常地大于已经被放大过的理论预测值。这意味着宇宙中的物质分布比单纯由我们的运动所能解释的更加不对称，宇宙可能天生就是歪斜的。

2021 年致力于推广开源软件和捍卫自由软件 GPL 许可证的非盈利组织 Software Freedom Conservancy(SFC) 对 Vizio 提起诉讼，指控其多次未能履行 GPL 许可证的基本要求。Vizio 是一家产品主要为高清电视机的消费电子品牌，创始人是王蔚。SFC 称 Vizio 电视机产品使用的 SmartCast 系统包含了 GPL 授权的软件，按照 GPL 许可证要求，购买 Vizio 产品的消费者有权访问源代码，允许对源代码进行修改、研究和在适当条件下重新发行。SFC 寻求 Vizio 履行其合规义务。现在法官做出一项裁决：GPLv2 许可证并不要求提供在设备上安装软件修改版本所需的签名密钥。Linus Torvalds 认为 GPLv2 许可证所要求的是提供源代码，针对的是软件，而不是扩大到硬件，要求获得硬件的访问权限。GPLv2 并没有强迫厂商开放硬件。Torvalds 批评（或者说抨击）了 SFC 的做法。

以易于使用著称、基于 Ubuntu 的发行版 elementary OS 释出了 v8.1 版本。主要变化包括：默认使用 Wayland 会话；改进窗口管理和多任务处理等。Elementary OS 8.1 支持 Arm64 设备，意味着用户可以在 Apple M 系列设备或其它支持加载 UEFI 固件的设备上运行 elementary OS。

尼安德特人如何消失或为何灭绝至今是一个争论的主题。有关尼安德特人灭绝的假说包括：人口下降、环境变化、与智人（即现代人类）竞争失败，以及基因同化。意大利和瑞士的研究人员在《Scientific Reports》上发表了一篇论文赞同基因同化的观点，认为尼安德特人是被现代人类吸收了，我们就是尼安德特人，不分彼此。尼安德特人部落的规模比智人的部落通常小一个数量级，当尼安德特人部落不断有智人迁入，反复的杂交和基因混合最终导致规模较小的尼安德特人被智人吸收。

Ötzi aka 冰人（Iceman）是一具因冰封而保存完好的天然木乃伊，距今有约 5000 年历史，其发现地点是阿尔卑斯山脉厄茨塔尔山冰川。Ötzi 以其保存完好的衣物、武器和纹身而知名，他的死因可能是肩膀上的箭头。他还被发现饱受骨折、肠道寄生虫和熏黑肺部的折磨。根据发表在 bioRxiv 上的一篇论文，科学家又发现了他的一项新疾病：致癌的人类乳头瘤病毒 HPV16。科学家报告 Ötzi 以及在西伯利亚西部发现的距今 4.5 万年的智人化石都携带了 HPV16 的 DNA 片段。HPV16  致癌病毒同时存在于相距 5000 公里相隔 4 万年的人类身上，显示它已经在人类中间传播了很长时间，可能是现代人类将其传播给了尼安德特人，而不是尼安德特人传播给人类。研究人员发现尼安德特人携带的是低风险的人类乳头瘤病毒 HPV12，而非高致癌性的 HPV16。最新发现挑战了 HPV 病毒是人类与尼安德特人杂交而感染的观点。

自由软件基金会（FSF）宣布收到了两笔总额约 90 万美元的捐赠。两笔捐款以门罗币(Monero) 形式捐出，是 FSF 迄今收到的金额最高的私人捐赠之一。捐赠者希望保持匿名。FSF 的资金主要来自个人捐赠和会员支持。两笔捐款让 FSF 提前实现了冬季筹款目标，它将把工作重点转向会员发展，目标是到 1 月 16 日发展 100 名准会员(associate member)。

最近发生多起发射事故的俄罗斯宇航局 Roscosmos 宣布计划到 2036 年在月球上建造一座核电站，已经与宇航公司 Lavochkin Association 签署了一份合同。Roscosmos 表示参与者还包括了国家原子能公司（Rosatom）和核能研究机构 Kurchatov Institute。Roscosmos 表示该电站将为俄罗斯的月球计划提供动力，包括月球车、天文台以及俄中联合国际月球研究站（Russian-Chinese International Lunar Research Station）的基础设施。

欧盟 2024 年使用的能量有 25.2% 来自可再生能源，比 2023 年增加 0.7 个百分点，距离 20230 年可再生能源占比 42.5% 还差 17.3 个百分点，意味着要实现目标从 2025 年到 2030 年可再生能源占比每年要增长 2.9%。欧盟国家中，瑞典的可再生能源占比最高达到 62.8%。瑞典主要依赖固体生物质能、水力发电和风能。芬兰紧随其后占比 52.1%，同样依赖固体生物质能、风能和水力发电。丹麦第三占比 46.8%，大部分可再生能源来自固体生物质能、风能和沼气。比利时（14.3%）、卢森堡（14.7%）和爱尔兰（16.1%）的可再生能源占比最低。

微软杰出工程师 Galen Hunt 在 LinkedIn 上畅谈要在 AI 和算法的帮助下，到 2030 年用 Rust 语言重写所有 C 和 C++ 代码，目标是一名工程师一个月一百万行代码。这番话引发了一片哗然，以至于微软进行澄清，而 Galen Hunt 则修改了他的帖子。他在帖子里大量使用“我们（We）”这个词，因此在外界看来他是代表公司发言。微软高管 Frank X. Shaw 澄清，公司没有计划使用 AI 重写 Windows 11。而 Galen Hunt 也在帖子里澄清：微软没有采用 AI 使用 Rust 重写 Windows 11，表示读者过度解读了。

Ruby 语言在圣诞节释出了 v4.0.0。Ruby 语言一直习惯在圣诞节发布大更新。Ruby 4.0.0 的新特性包括：新实验性功能 Ruby Box——提供定义隔离；新的 JIT 编译器 ZJIT，它是作为 YJIT 的下一代开发的，但目前速度还不如 YJIT，不建议用于生产环境；改进并行执行机制 Ractor；语法方面的改变，等等。

在获得出口许可之后，英伟达通知中国客户计划春节前发货 H200。初始订单将使用现有库存完成，现有库存有 5,000-10,000 块 HGX 主板，总计提供 40,000 到 80,000 块 GPU。这意味着英伟达将优先供应性能更强的 SXM 版的 H200 显卡，比基于 PCIe 的 NVL 显卡更适合训练应用。根据英伟达与美国政府的协议，它将上缴 25% 的销售收入。英伟达同时通知客户，在中国政府批准之后才能确定发货时间。