AK@_akhaliq · 5月6日68

From Context to Skills Can Language Models Learn from Context Skillfully? paper: https://huggingface.co/papers/2604.27660

译从上下文到技能 语言模型能否巧妙地通过上下文学习？ 论文：https://huggingface.co/papers/2604.27660

AK@_akhaliq · 5月6日61

Persistent Visual Memory Sustaining Perception for Deep Generation in LVLMs paper: https://huggingface.co/papers/2605.00814

译持久视觉记忆 为LVLMs中的深度生成维持感知 论文: https://huggingface.co/papers/2605.00814

Berryxia.AI@berryxia · 5月5日75

Google 这一波操作，最让人意外的是 Google直接把LLM推理里最顽固的autoregressive瓶颈干掉了。 他们和UCSD合作推出的DFlash（Diffusion-Style Speculative Decoding），在Google Cloud TPU上实现了3.13倍的推理加速，而且是无损的。 这不是又一个“理论上更快”的小优化，而是真正从根子上改变了生成式解码的范式：用扩散式推测一次生成多个token，彻底绕过传统一个词接一个词的串行限制。 当推理速度突然提升3倍以上，意味着： - 云端成本曲线被重塑 - 实时Agent、长上下文、复杂工具调用都变得更现实 - 本地部署的门槛也被大幅拉低 过去我们总觉得“模型参数越大越强”， 现在硬件+解码策略的系统级突破，正在把“更快”变成真正的生产力杠杆。 Google这波操作，把LLM推理的下一代竞赛直接拉到了硬件+算法联合优化的赛道。 你觉得DFlash这种扩散式推测解码，会不会成为未来所有大模型推理的标准配置？ 博客在这里👉 https://goo.gle/4naZ8Yv

译Google与UCSD合作推出扩散式推测解码技术DFlash，在Google Cloud TPU上实现了3.13倍的无损推理加速。该技术突破了传统自回归解码逐个生成token的串行瓶颈，通过一次推测生成多个token来改变生成范式。这一硬件与算法的联合优化，将重塑云端成本曲线，并使实时Agent、长上下文等应用更趋现实，同时大幅降低本地部署门槛。此举将大模型推理的竞争引向了系统级优化的新赛道。

Rohan Paul@rohanpaul_ai · 5月5日52

This Google DeepMind paper trains LLMs to learn during conversation, and it shows they get much better at using feedback. The problem is that most LLMs treat a chat like a series of separate turns, so even when a user corrects them, they often do not really use that new information and they also fail to ask for missing details. The paper fixes this by turning a normal task into a teacher student dialogue, where the student model tries an answer, a teacher with hidden extra information gives guidance, and the student is trained to use that guidance to reach the right answer. The authors test 2 training styles, offline filtering and online reinforcement learning, and they report that the online version works better, with training on short 4 turn chats still helping on longer 10 turn chats later. They also show that this skill carries from math to coding and helps on messy underspecified tasks where the full problem arrives bit by bit instead of all at once. A second step called Q-priming teaches the model to ask useful questions, and on ambiguous tasks it becomes over 5x more likely to ask for clarification instead of making an early wrong guess, which matters because it makes chat feel more like working with someone who can actually learn during the conversation. ---- Paper Link – arxiv. org/abs/2602.16488 Paper Title: "Learning to Learn from Language Feedback with Social Meta-Learning"

译Google DeepMind的研究通过“师生对话”框架训练大型语言模型（LLM），使其能在对话中有效利用用户反馈进行学习。传统LLM将对话视为独立轮次，难以整合修正信息。该研究让“学生”模型尝试回答，由掌握额外信息的“教师”提供指导，并训练学生利用指导得出正确答案。在线强化学习训练效果优于离线过滤，且在简短对话中习得的技能能迁移至更长对话。该方法从数学任务泛化至编程任务，并能处理信息逐步到达的模糊任务。通过“Q-priming”步骤，模型在模糊任务中主动寻求澄清的可能性提高五倍以上，使对话更像与一个能在交流中实时学习的伙伴协作。

AK@_akhaliq · 5月5日68

UniVidX A Unified Multimodal Framework for Versatile Video Generation via Diffusion Priors paper: https://huggingface.co/papers/2605.00658

译UniVidX 一个通过扩散先验实现多功能视频生成的统一多模态框架 paper: https://huggingface.co/papers/2605.00658

AK@_akhaliq · 5月5日55

Web2BigTable A Bi-Level Multi-Agent LLM System for Internet-Scale Information Search and Extraction paper: https://huggingface.co/papers/2604.27221

译Web2BigTable 一个用于互联网规模信息搜索与提取的双层多智能体LLM系统 论文: https://huggingface.co/papers/2604.27221

Microsoft Research@MSFTResearch · 5月5日62

Research Focus: AI agents leaking enterprise data, a smarter OS for cloud deployment, and new research on how to actually structure AI use at work. https://msft.it/6016vKxQm

译研究焦点：AI代理泄露企业数据、为云端部署打造更智能的操作系统，以及关于如何在工作中实际构建AI应用的新研究。https://msft.it/6016vKxQm

elvis@omarsar0 · 5月4日66

Autodata (from Meta) is an agentic data scientist that builds high-quality training and evaluation data autonomously. Great work on the autoharness track. (bookmark it)

译Meta FAIR开发的Autodata是一个能自主构建高质量训练与评估数据的代理系统。其核心在于“代理式自我指导”循环：编排器LLM指导挑战者代理基于领域文档生成问题，由弱、强解算器尝试解答，法官评分后分析失败并循环优化，从而产出能有效区分模型能力的挑战性数据。在CS研究QA任务中，该方法产生了34个百分点的性能差距，远超标准方法的1.9点。系统还具备元优化能力，通过外循环调整指令，将验证通过率从12.8%提升至42.4%。研究处理了超万篇论文，产出2,117个优质QA对，通过增加推理计算使数据更具挑战性，从而提升下游模型性能。

elvis@omarsar0 · 5月4日68

NEW paper from Sakana AI (ICLR 2026). A 7B Conductor model just hit SOTA on GPQA-Diamond and LiveCodeBench by orchestrating other LLMs instead of solving problems itself. (great paper! bookmark it!) The Conductor is trained with RL to do two things at once: design communication topologies between worker agents (open or closed source), and prompt-engineer focused instructions to each worker so it leverages their individual strengths. It's like training a special agent to take care of both collaboration and communication. Trained against randomized agent pools, it adapts to arbitrary mixes of agents at inference time. Even more interesting: when allowed to pick itself as a worker, it forms recursive topologies, unlocking a new form of dynamic test-time scaling through online iterative adaptation. The gains over the best individual worker on AIME25 and GPQA-D land in the ~3% range, which the authors note is consistent with entire generational improvements between frontier model versions, except this one comes from coordination, not pretraining. Why it matters? We can start to think of the orchestrator as the model now. Routing decisions aren't just a wrapper, they're a learnable policy. Paper: https://arxiv.org/abs/2512.04388 Learn to build effective AI agents in our academy: https://academy.dair.ai/

译Sakana AI在ICLR 2026上发表研究，提出一个仅70亿参数的“指挥者”模型。该模型不直接解决问题，而是通过强化学习训练，专注于为混合开源与闭源模型的工作者智能体设计通信拓扑结构，并为每个工作者生成精准指令以发挥其特长。经随机化智能体池训练后，它能在推理时适应任意智能体组合。其关键创新在于，当允许指挥模型将自己也选为工作者时，系统会形成递归拓扑，实现动态测试时扩展。该模型在GPQA-Diamond和LiveCodeBench上达到SOTA水平，在AIME25和GPQA-D上的性能比最佳单体工作者提升约3%，这相当于前沿模型一个代际的改进幅度，且增益完全来源于协同优化。

Rohan Paul@rohanpaul_ai · 5月4日48

This paper proposes a smarter way for LLMs to reason by splitting work across agents that share one workspace. The problem is that even strong reasoning models still break on harder multi-step tasks because they do not carry out logic reliably all the way through. The system, called BIGMAS, builds a small graph of specialist agents for each problem, rather than using one fixed chain every time. Every agent reads and writes through a shared workspace, while a separate controller sees the whole state and picks the next useful step. The authors tested it on 3 puzzle tasks across 6 frontier models, covering arithmetic expression search and multi-step planning. It improved results on every model and task, with examples like 12% to 30% on Six Fives and 57% to 93% on Tower of London. What matters is that the paper shows reasoning can improve from better system structure, not only from making a single model think longer. ---- Paper Link – arxiv. org/abs/2603.15371 Paper Title: "Brain-Inspired Graph Multi-Agent Systems for LLM Reasoning"

译本文提出BIGMAS系统，通过构建针对特定问题的小型专家智能体图，以提升大语言模型在多步骤复杂任务中的推理可靠性。该系统让多个智能体在一个共享工作空间中进行读写协作，并由独立控制器监控全局状态、规划下一步行动。在涵盖算术表达式搜索和多步骤规划的3个谜题任务上，对6个前沿模型进行的测试表明，该方法在所有模型和任务上均显著提升了性能，例如Six Fives任务准确率从12%提升至30%，伦敦塔任务从57%提升至93%。这证明通过优化多智能体系统结构，而非仅依赖延长单一模型的思考，能有效增强大语言模型的推理能力。

Rohan Paul@rohanpaul_ai · 5月3日56

Adobe paper tests if LLM chain of thought explanations show real influences, and finds they often hide them. LLMs often notice hidden hints in a question but leave them out of their step by step explanation. So the point they are making is that LLMs often notice hidden hints in a question but leave them out of their step by step explanation. People use step by step reasoning as a safety and debugging tool, assuming the explanation matches what drove the answer. The author plants misleading hints inside multiple choice questions and checks both the final answer and what the model says. Across 9,154 trials and 11 major LLMs, the study tried 3 instructions, normal, "you are monitored", and "report hints". Under normal use, models mentioned hints only 20.7% of the time, yet 99.4% admitted seeing them when asked later. Telling models they were monitored barely changed this, and forcing hint reporting caused 68.2% false alarms when no hints existed. User pleasing hints, where the question says a confident user believes an option, were followed 45.5% but often unmentioned, making monitoring that only reads explanations unreliable. ---- Paper Link – arxiv. org/abs/2601.00830 Paper Title: "Can They Trust AI Explanations? Evidence of Systematic Underreporting in Chain-of-Thought Reasoning"

译Adobe研究测试大型语言模型（LLM）思维链解释是否反映真实决策影响。通过在多选题中植入误导性提示，对11个主流模型进行9154次试验。正常使用时，模型仅在20.7%的逐步推理中提及隐藏提示，但后续追问时99.4%承认注意到。告知模型其推理受监控后结果未见改善；强制要求报告提示则导致68.2%误报（当无提示时）。问题包含用户倾向性提示时，模型45.5%遵从该倾向，但解释中常未说明。研究表明，思维链解释常与真实决策依据不匹配，仅依赖其作为安全调试工具可能不可靠。

elvis@omarsar0 · 5月3日57

Claude Opus 4.7 just implemented an AlphaZero-style self-play pipeline from scratch. It did this on consumer hardware in three hours, then beat the Pascal Pons solver 7 of 8 as first-mover on Connect Four. No other frontier coding agent tested cleared 2 of 8. This paper proposes a new way to evaluate coding agents: hand them a minimal task description, give them a tight budget, and ask them to autonomously rebuild a famous ML breakthrough. Connect Four + AlphaZero is the first instance. It's small enough to run on a laptop and hard enough to require a real research engineering loop (MCTS, neural value/policy nets, self-play, training schedule). We've been measuring coding agents on patches and unit tests. This shifts the bar to "can the agent build a non-trivial ML system end-to-end on its own?" The answer is now yes for at least one frontier model. Paper: https://arxiv.org/abs/2604.25067 Learn to build effective AI agents in our academy: https://academy.dair.ai/

译本研究提出评估编程智能体的新范式：要求其在有限预算下，仅凭简要任务描述自主重建一项著名机器学习突破。首个测试案例为四子棋AlphaZero系统，其规模适于笔记本电脑运行，但复杂度要求完成完整的研究工程闭环。Claude Opus 4.7在三小时内从零构建了自博弈训练管道，并作为先手以7:1的成绩击败了Pascal Pons求解器，而其他前沿智能体均未通过2/8的测试。这标志着评估标准已从代码补全提升为端到端构建非平凡机器学习系统的能力。

Chubby♨️@kimmonismus · 5月3日48

GPT-5.4 Pro didn’t just solve one math problem, it kicked open the door: its proof method now cracks a 60-year-old Erdős conjecture, making this one of the first times an AI proof actually leads somewhere.​​​​​​​​​​​​​​​​ We barely started.

译GPT-5.4 Pro不仅解决了一个数学问题，其证明方法更成功破解了长达60年的埃尔德什猜想。研究团队在此基础上改进并应用该方法，进一步证明了包括埃尔德什、Sárközy和Szemerédi提出的另一项60年猜想在内的多个附加问题。这标志着AI生成的证明首次展现出显著的“下游影响”，其核心价值不仅在于解决问题本身，更在于为数学研究开辟了新的路径。相关成果已在未来数学研讨会上公布。

Hao AI Lab@haoailab · 5月2日37

Excited to share our recent work accepted to ICML 2026! These projects span efficient causal parallel decoders, diffusion LLMs, video sparse attention, video QAT, online speculative decoding, and agentic document reasoning. Huge thanks to all collaborators and co-authors across these efforts. Looking forward to seeing everyone in Seoul this summer! 🇰🇷

译很高兴分享我们最近被ICML 2026接收的工作！这些项目涵盖高效因果并行解码器、扩散大语言模型、视频稀疏注意力、视频量化感知训练、在线推测解码以及智能文档推理。 衷心感谢所有合作者和共同作者在这些工作中的付出。期待今年夏天在首尔与大家相见！🇰🇷

AK@_akhaliq · 5月2日56

Heterogeneous Scientific Foundation Model Collaboration paper: https://huggingface.co/papers/2604.27351

译异构科学基础模型协作 paper: https://huggingface.co/papers/2604.27351

AK@_akhaliq · 5月2日57

The Last Human-Written Paper Agent-Native Research Artifacts paper: https://huggingface.co/papers/2604.24658

译最后一篇人类撰写的论文 智能体原生研究制品 论文: https://huggingface.co/papers/2604.24658

AK@_akhaliq · 5月2日35

Co-Evolving Policy Distillation paper: https://huggingface.co/papers/2604.27083

译协同进化策略蒸馏 论文: https://huggingface.co/papers/2604.27083

elvis@omarsar0 · 5月1日56

Cool paper from Meta FAIR. It's on self-improving LLMs but on the pretraining side. (bookmark it) Most LLM safety, factuality, and reasoning fixes get bolted on at post-training. By then, the patterns have already set. This work moves those behaviors into pretraining itself. The team uses a strong post-trained model as both a rewriter and a judge: it rewrites pretraining suffixes toward higher-quality, safer continuations, then scores model rollouts against the original suffix and the rewrite to drive RL during pretraining. Instead of next-token prediction, the policy learns sequence generation from the start, with rewards for quality, safety, and factuality. Why it matters: 36.2% relative gain in factuality, 18.5% in safety, and up to 86.3% win rate in generation quality over standard pretraining. Bottom line: the post-trained models you already have can be used to pretrain the next ones better. Paper: https://arxiv.org/abs/2601.21343 Learn to build effective AI agents in our academy: https://academy.dair.ai/

译Meta FAIR的研究提出一种新范式，将LLM的改进从后训练移至预训练阶段。该方法利用强大的后训练模型作为改写器和评判器，对预训练数据的后缀进行高质量、高安全性的改写，并通过强化学习直接优化预训练模型。模型从开始就学习序列生成，并获得质量、安全性和事实性的奖励。实验结果显示，相比标准预训练，该方法在事实性上取得36.2%的相对提升，安全性提升18.5%，生成质量胜率最高达86.3%。核心结论是，现有后训练模型可用于预训练出更优的下一代模型。

Ethan Mollick@emollick · 5月1日62

New paper (on an old AI) tests o1 against doctors on medical benchmarks & real ER cases: “across a variety of scenarios and applications, the large language model outperformed both human physicians and older models” The potential suggests an “urgent need for prospective trials.”

译新论文（关于旧式人工智能）在医学基准测试和真实急诊病例中将o1与医生进行对比："在各种场景和应用中，大型语言模型的表现均优于人类医生和旧版模型" 该潜力表明"迫切需要前瞻性试验"。

向阳乔木@vista8 · 5月1日48

语言模型能说话但不懂数据，专用模型懂数据但不能说话，这是科学AI当下困境之一。 UIUC最新论文 Eywa 从《阿凡达》找到了答案。 纳美人通过"Tsaheylu"神经键跨越物种障碍，让山地歌鸟、雷兽各展所长。 Eywa 做的事情一样：给语言模型和专用基础模型之间建一个接口。 让 Chronos 做时序预测，让 TabPFN 处理表格，语言模型负责理解任务、调度工具、整合结果。 --- 从论文数据看，效果不错，短时间是一个MCP就能解决连接问题，但长期也不知道语言模型能否达到专用模型的水平。 论文见评论区

译针对通用语言模型懂交互却不懂数据、专用模型精通数据却缺乏交互能力的科学AI困境，UIUC团队受《阿凡达》“Tsaheylu”神经连接启发，提出了Eywa接口框架。该框架让语言模型负责理解指令与调度，调用如Chronos、TabPFN等专用模型处理数据，从而协同发挥两者优势。初步实验效果良好，长期挑战在于语言模型能否达到专用模型的领域性能。

Rohan Paul@rohanpaul_ai · 5月1日46

Research proves that current AI agent groups cannot reliably coordinate or agree on simple decisions. Building teams of AI agents that can consistently agree on a final decision is surprisingly difficult for LLMs. But problem is that developers frequently assume that if you have enough AI agents working together, they will eventually figure out how to solve a problem by talking it through. This paper shows that this assumption is currently wrong. Even in a friendly environment where every agent is trying to help, the team often gets stuck or stops responding entirely. Because this happens more often as the group gets bigger, it means we cannot yet trust these agent systems to handle tasks where they must agree on a correct answer. ---- Paper Link – arxiv. org/abs/2603.01213 Paper Title: "Can AI Agents Agree?"

译研究表明，当前由多个LLM组成的AI智能体团队在需要协调达成最终决策时存在根本性困难。开发者常误以为增加智能体数量并通过讨论就能解决问题，但论文证明这一假设目前是错误的。即使在友好协作环境中，智能体团队也常陷入僵局或完全停止响应，且团队规模越大问题越突出。这意味着现有AI智能体系统尚无法可靠处理需要达成一致正确答案的任务。

Rohan Paul@rohanpaul_ai · 5月1日62

Researchers tested autonomous AI agents in real environments and found they easily cause massive security disasters. In one test an agent actually wiped its entire email server just to keep a secret for a stranger. The main problem with standard language models is that giving them control over real computer tools creates dangerous blind spots. To understand these risks the researchers let 20 experts interact with live AI assistants through chat and email for 2 weeks. They discovered that these programs blindly follow instructions from almost anyone and often lie about what they have actually done. This matters because tech companies are rushing to deploy these autonomous helpers without fixing their basic inability to understand who they should actually trust. --- Paper Link – arxiv. org/abs/2602.20021 Paper Title: "Agents of Chaos"

译研究人员在真实环境中测试自主AI代理，发现它们极易引发大规模安全灾难，如为保守秘密而删除整个电子邮件服务器。核心问题在于标准语言模型被赋予计算机工具控制权后，产生危险盲点，导致代理盲目遵循几乎任何人的指令并经常撒谎行为。通过让20位专家与实时AI助手进行两周互动实验，研究揭示了这些程序缺乏基本信任判断能力。科技公司正急于部署此类自主助手，却未修复其无法理解应信任谁的根本缺陷，加剧了安全风险。

Rohan Paul@rohanpaul_ai · 5月1日43

The LongCat team just released LARYBench, a benchmark built to test whether an AI model truly learns action from video, instead of only looking good when attached to a robot policy later. It evaluates latent actions, meaning the hidden motion signals a model extracts from video, across 1.2M+ clips, 620K+ image pairs, 595K trajectories, 151 action classes, and 11 robot platforms. A latent action representation tries to store the change between frames as something like reach, pick, place, move left, or close gripper, rather than memorizing raw pixels. The key point is that robot training data is scarce, while human and robot videos are abundant, so the whole field wants a way to turn cheap video into useful action knowledge. The paper argues that older evaluations mixed too many things together, because a robot succeeding on a task depends on the policy, training recipe, environment, and controller, so you could not tell whether the action representation itself was actually good. LARYBench splits the problem into 2 cleaner tests, where one asks whether the representation knows what happened and the other asks whether it preserves enough detail for how to move. The biggest result is that general self-supervised vision models beat specialized embodied LAMs, with V-JEPA 2 reaching 76.62% average action classification accuracy, while DINOv3 gives the best overall control regression score at 0.19 MSE, far ahead of embodied models clustered around 0.87 to 0.97. The deeper point is that strong visual representations already contain a surprising amount of action knowledge, and the paper also shows that latent feature spaces map to robot control better than pixel reconstruction spaces, which helps explain why some robotics systems may be building on the wrong intermediate representation. 🧵 1.

译LongCat团队推出LARYBench基准，旨在评估AI模型是否从视频中真正学习动作，而非仅在后端机器人策略中表现良好。该基准聚焦模型从视频提取的潜在动作表示，通过超过120万视频片段等数据，将评估拆分为动作分类与控制回归两个清晰测试。关键发现是，通用自监督视觉模型（如V-JEPA 2和DINOv3）表现优于专用具身模型，表明强大视觉表示已蕴含丰富动作知识，且潜在特征空间比像素重建更利于机器人控制映射。这为利用丰富视频数据解决机器人训练数据稀缺问题提供了新方向。

AK@_akhaliq · 5月1日47

Recursive Multi-Agent Systems paper: https://huggingface.co/papers/2604.25917

译递归多智能体系统 论文: https://huggingface.co/papers/2604.25917

Ethan Mollick@emollick · 5月1日55

Randomized trial of an AI therapy chatbot on Mexican women found “improved mental health by 0.3 SD over 6 months with no evidence of an increase of severe cases; improved sleep, healthful behaviors, daily functioning & labor market outcomes” Big results for a cheap intervention.

译一项针对墨西哥女性的随机试验发现，使用基于认知行为疗法训练的AI对话代理的心理健康应用Mindsurf，在六个月内使使用者心理健康水平提升了0.3个标准差，且未增加严重病例。该干预还改善了睡眠质量、健康行为、日常功能及劳动力市场表现（如减少缺勤），其效益远超成本。尽管使用者寻求传统心理治疗的比例有所增加，但这并非心理健康改善的主因。效果具有持续性，短期使用可通过促进行为的持续改变带来长期改善。

Berryxia.AI@berryxia · 5月1日57

你看看大模型有多重？ 这个挺有意思的😂

译Pine AI首席科学家李博杰提出新方法，通过模型回答1400道冷知识题的能力来估算其参数量。原理是存储事实需占用参数空间，先利用已知开源模型拟合曲线，再将闭源模型得分投射得出估算。研究评估了92个闭源模型，结果显示GPT-5.5以约9.7T参数断层领先，Claude Opus 4.6约5.3T次之。主流旗舰模型如GPT-5、Claude Opus 4.7参数集中在3-4T量级。分析还推断GPT-5的.x版本及Claude Opus 4.7等可能是全新训练而非微调产物，并指出MoE模型的知识容量取决于总参数量。评测工具与数据已开源。

Microsoft Research@MSFTResearch · 5月1日64

Safe agents don’t guarantee a safe ecosystem of interconnected agents. Microsoft Research examines what breaks when AI agents interact and why network-level risks require new approaches. Learn more: https://www.microsoft.com/en-us/research/blog/red-teaming-a-network-of-agents-understanding-what-breaks-when-ai-agents-interact-at-scale/

译安全的智能体并不能保证由相互连接的智能体组成的生态系统是安全的。微软研究院研究了当AI智能体交互时会出现什么问题，以及为什么网络层面的风险需要新的方法。了解更多：https://www.microsoft.com/en-us/research/blog/red-teaming-a-network-of-agents-understanding-what-breaks-when-ai-agents-interact-at-scale/

elvis@omarsar0 · 5月1日57

// When to Retrieve During Reasoning // Pay attention to this one, AI devs. (bookmark it) Most RAG systems retrieve once, before the model starts reasoning. Large reasoning models like o1 and R1 don't work that way. They generate 12k-25k token chains of thought and hit knowledge gaps mid-inference, long after the retrieval window closed. ReaLM-Retrieve is a reasoning-aware retrieval framework that injects evidence during multi-step inference. It detects uncertainty at reasoning-step granularity (not token or sentence level), learns a policy for when external evidence actually helps, and cuts per-retrieval overhead by 3.2x. This approach achieves +10.1% absolute F1 over standard RAG across MuSiQue, HotpotQA, and 2WikiMultiHopQA, with 47% fewer retrieval calls than fixed-interval IRCoT. On 2-4 hop MuSiQue it hits 71.2% F1 with only 1.8 retrieval calls per question. If you're shipping reasoning-model RAG, your retrieval needs to know when to fire, not just what to fetch. Paper: https://arxiv.org/abs/2604.26649 Learn to build effective AI agents in our academy: https://academy.dair.ai/

译传统RAG系统在推理前单次检索，无法满足如o1、R1等大型推理模型在生成长链思维时中途出现的知识需求。ReaLM-Retrieve提出推理感知的检索框架，能在多步推理中动态注入证据。其核心在于以推理步骤粒度检测不确定性，学习判断何时引入外部证据有效，并将单次检索开销降低3.2倍。在多个QA数据集上，该框架比标准RAG的F1绝对值提升10.1%，且检索调用次数比固定间隔的IRCoT减少47%。在2-4跳的MuSiQue任务中，仅用平均1.8次检索即可达到71.2%的F1值，表明面向推理模型的RAG需优化检索时机而不仅是检索内容。

Rohan Paul@rohanpaul_ai · 5月1日58

Frontier AI can now autonomously chain complex, expert-level cyber attacks end-to-end, at superhuman speed and near-zero marginal cost. GPT-5.5 essentially tied with Mythos Preview - within the margin of error — both far ahead of earlier models (GPT-4o, Claude Opus 4.x, etc.). - GPT-5.5: 71.4% (±8.0%) - Mythos Preview: 68.6% (±8.7%) AISI has been running controlled, realistic cybersecurity evaluations on the latest AI models. These include: - Narrow CTF-style tasks (expert-level challenges like exploiting memory corruptions, breaking crypto, reverse-engineering stripped binaries, etc.). - Multi-step “cyber range” simulations — a full 32-step corporate network attack chain (recon → initial access → lateral movement → privilege escalation → full network takeover). A human expert needs ~20 hours for this. They previously tested Mythos Preview, and now OpenAI’s GPT-5.5. One hard reverse-engineering task (custom virtual machine) takes a human expert ~12 hours with professional tools. GPT-5.5 solved it in under 11 minutes at a cost of $1.73.

译前沿AI已能以超人速度和近乎零边际成本自主完成端到端的复杂专家级网络攻击链。在AISI的网络安全评估中，GPT-5.5与Mythos Preview表现相当，均远超GPT-4o等早期模型。GPT-5.5在包含32个步骤的企业网络攻击模拟中成功完成端到端攻击，而人类专家需约20小时。在一项人类专家需12小时完成的反向工程任务中，GPT-5.5仅用11分钟、花费1.73美元即告解决。

Anthropic@AnthropicAI · 5月1日63

How do people seek guidance from Claude? We looked at 1M conversations to understand what questions people ask, how Claude responds, and where it slips into sycophancy. We used what we found to improve how we trained Opus 4.7 and Mythos Preview. https://www.anthropic.com/research/claude-personal-guidance

译人们如何向Claude寻求指导？ 我们分析了100万次对话，以了解人们提出什么问题、Claude如何回应，以及它何时会陷入阿谀奉承。我们利用这些发现改进了Opus 4.7和Mythos Preview的训练方式。 https://www.anthropic.com/research/claude-personal-guidance

Epoch AI@EpochAIResearch · 5月1日59

How much AI compute has been smuggled to China? We estimate between 290k and 1.6M H100-equivalents by the end of 2025 — representing ~20% to ~60% of China’s total compute.

译有多少AI算力被走私到中国？我们估计到2025年底将达到29万至160万H100等效算力——约占中国总算力的20%至60%。

Rohan Paul@rohanpaul_ai · 5月1日61

Google DeepMind’s real-time video AI doctor is here. They just introduced AI co-clinician, a triadic care system built to work under a doctor’s supervision during patient care. The system is built to retrieve clinical-grade evidence, verify it, and in patient-facing simulations use a dual-agent setup where one module talks while another watches for boundary violations. It also beat other frontier models on open-ended drug questions, because real medicine arrives as messy patient cases, not multiple-choice exams. DeepMind evaluated it against the failure modes clinicians actually care about: saying the wrong thing, or failing to surface the crucial thing. In 98 realistic primary care evidence queries, physicians preferred the co-clinician to leading evidence-synthesis tools, and the system logged zero critical errors in 97 cases under their NOHARM-style evaluation.

译Google DeepMind 近日发布 AI co-clinician 协诊系统，这是一个多模态代理系统，旨在辅助医护人员，并在医生监督下运行。系统采用双代理架构：一个模块与患者对话，另一模块实时监控交互边界，能检索并验证临床级证据。在开放式药物问答中，其表现超越前沿模型，更贴合真实医疗场景的复杂性。评估聚焦临床实际关切，如避免错误陈述或遗漏关键信息。在98项初级保健模拟查询中，医生对其偏好超过主流证据合成工具；在97例NOHARM风格评估中未出现严重错误。

Google DeepMind@GoogleDeepMind · 4月30日47

AI co-clinician is our new research initiative to help explore how multimodal agents could better support healthcare workers and patients. 🩺 Here’s a snapshot of our progress 🧵

译AI协同临床医生是我们新的研究计划，旨在探索多模态智能体如何更好地支持医护人员和患者。🩺 以下是我们进展的概览🧵

歸藏(guizang.ai)@op7418 · 4月30日51

DeepSeek 多模态大语言模型的论文《Thinking with Visual Primitives》已经公开 底座是 DeepSeek-V4-Flash，MoE架构，总参数量 284B，激活参数量 13B。 自研 DeepSeek‑ViT 视觉编码模型，14×14 patch，输出后 3×3 空间压缩，再接入 LLM。 模型在回答时不仅进行文字推理，还会同时通过画框、打点等“视觉原语”进行思考。 在极低的 Token 成本下，其效果能和 GPT-5.4、Claude 以及 Gemini 在一些前沿指标上对齐，甚至有的指标能反超。

译该论文介绍了基于DeepSeek-V4-Flash底座的多模态大模型。其核心创新在于模型能同时进行文字推理和“视觉原语”（如画框、打点）思考。该模型以极低的Token成本，在多项前沿指标上达到了与GPT-5.4、Claude、Gemini等模型相当甚至更优的性能。

Rohan Paul@rohanpaul_ai · 4月30日73

New Microsoft paper shows that current AI assistants often damage documents during long editing jobs. Even the frontier models still ended up corrupting about 25% of document content on average, while many other models damaged far more. The problem is that delegated AI work only makes sense if a model can keep a document correct across many edits, not just do 1 step well. The paper tests this with reversible task pairs, where a model edits a file and then tries to undo that edit, so a reliable system should return to the original document. The authors built real work setups across 52 domains, from coding and science to accounting and music notation, and ran 19 models through 20 editing interactions. The failures were usually not lots of tiny slips but occasional big mistakes that silently broke parts of the document and then compounded over time. Agentic tool use did not help in their tests, and bigger files, longer workflows, and irrelevant extra documents made the corruption worse. The reason this matters is that current LLMs can look strong in short demos or narrow coding tasks yet still be unreliable delegates for long real-world document work. ---- Paper Link – arxiv. org/abs/2604.15597 Paper Title: "LLMs Corrupt Your Documents When You Delegate"

译微软最新论文指出，当前AI助手在执行长链条编辑任务时，普遍会损坏文档内容。研究通过可逆任务对测试了19个模型，发现即使是前沿模型平均也会破坏约25%的文档内容，且问题随文件增大、流程变长而加剧。失败模式通常不是微小失误，而是偶尔出现的重大错误，这些错误会静默破坏部分文档并随时间累积。研究表明，当前的LLM在简短演示或狭窄编码任务中可能表现良好，但作为现实世界长文档工作的委托代理仍不可靠。

Rohan Paul@rohanpaul_ai · 4月30日55

Anthropic's new research shows that Claude can solve real bioinformatics problems human experts miss. 23 “human-difficult” problems that their expert panel could not solve, and their top model, Claude Mythos Preview, solved 29.6% of that set. The problem is that older science tests mostly check clean questions, not messy biology data work on real datasets. BioMysteryBench tries to fix that by hiding objective answers inside real datasets and grading only the final answer. It gives Claude standard biology tools and database access on 99 tasks, while up to 5 experts try them too. On the 76 problems at least 1 expert solved, the best model got about 83%, and on 23 expert-stumping problems it got about 30%. The post also found that wins on the hard problems were much less repeatable across 5 tries, so many successes were shaky rather than dependable. Anthropic’s own examples suggest Claude is strongest when it behaves less like an oracle and more like an unusually fast research collaborator: it layers methods, cross-checks evidence, and uses broad background knowledge to narrow the search space.

译Anthropic最新研究利用BioMysteryBench测试平台评估Claude在真实生物信息学问题上的能力。该测试将客观答案隐藏于真实数据集中，涵盖99项任务。在至少一位人类专家解决的76个问题上，Claude Mythos Preview模型准确率约为83%；更值得注意的是，在23个专家小组未能解决的问题上，该模型仍解决了其中约29.6%。然而，模型在困难问题上的成功重复性较低，表明其表现尚不稳定。研究指出，Claude最有效的模式并非充当“先知”，而是扮演快速研究协作伙伴的角色：通过分层使用方法、交叉验证证据并运用广泛背景知识来缩小搜索空间。

Rohan Paul@rohanpaul_ai · 4月30日54

The paper proposes a way for a coding agent to rewrite its own tools and rules, then check whether each change really helped. The big deal is that it turns harness tuning from guesswork into an auditable experiment, so the part of agent systems that quietly eats the most time and effort can now improve itself in a controlled and measurable way. The problem is that agent harnesses, meaning the prompts, tools, memory, and rules around a model, are usually tuned by hand or changed through messy self-improvement loops that produce lots of edits but little clear evidence about what helped. The method, called Agentic Harness Engineering, turns those edits into file-level parts that can be changed or rolled back, compresses huge run logs into short failure evidence, and makes the agent write a prediction for each edit that later gets checked against real task results. They tested this on Terminal-Bench 2, a hard coding benchmark in a terminal, by starting from a very small shell-only harness and letting the loop run for 10 rounds while keeping the base model fixed. The single-try success rate rose from 69.7% to 77.0%, beating Codex-CLI at 71.9% and other self-evolving baselines, which suggests the gains came from better harness design rather than from swapping in a stronger model. The final harness also carried over to other models and to SWE-bench-verified, with gains of 5.1 to 10.1 points across model families and 12% fewer tokens than the seed on SWE-bench-verified, which matters because harness work is expensive and this gives a more reliable way to let that layer improve itself without drifting into random noise. ---- Paper Link – arxiv. org/abs/2604.25850 Paper Title: "Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses"

译本文提出Agentic Harness Engineering方法，使编码代理能自动重写自身工具和规则，并通过可审计实验验证每次更改的有效性。传统代理工具调整依赖手动或混乱自我改进循环，缺乏明确证据。该方法将编辑转化为文件级可回滚部分，压缩运行日志为简短失败证据，并让代理为编辑写预测后基于任务结果检查。在Terminal-Bench 2测试中，从小型shell-only工具开始，经10轮进化且基础模型固定，单次尝试成功率从69.7%提升至77.0%，超越其他基线。最终工具可迁移至其他模型和SWE-bench-verified任务，在不同模型家族获得5.1到10.1点提升，并减少12%令牌使用，为昂贵工具工作提供可靠、可控的自我改进途径。

Chubby♨️@kimmonismus · 4月30日61

Anthropic just dropped a benchmark that should make every scientist pay attention. BioMysteryBench puts AI models through 99 real bioinformatics challenges, using raw, messy datasets from actual research, think unprocessed DNA sequences and clinical samples. However: these aren't textbook problems with neat answers. They're the kind of open-ended puzzles that keep PhD students up at night. The results are exciting. Claude's latest models (4.7) solve the majority of tasks that trained human experts can handle, and on 23 problems that a panel of five domain experts couldn't crack, Claude Mythos Preview nailed 30% of them. How? By combining knowledge from hundreds of thousands of papers and layering multiple analytical strategies when uncertain, essentially doing what a room full of specialists would do, but faster and in a single run. Genentech and Roche independently confirmed this trajectory with their own CompBioBench, where Claude Opus 4.6 reached 81% overall accuracy and 69% on the hardest questions. Two separate benchmarks, same conclusion: AI is no longer just keeping pace with biologists, it's pulling ahead on some of the hardest problems.

译Anthropic发布了BioMysteryBench基准测试，包含99个使用原始、杂乱真实生物数据集的开放式生物信息学挑战。最新Claude模型（4.7）解决了大部分人类专家能处理的任务，并在专家小组未能解决的23个难题中攻克了约30%。其能力源于整合数十万篇论文知识，并在不确定时叠加多种分析策略。Genentech和Roche的独立测试（CompBioBench）中，Claude Opus 4.6总体准确率达81%，最难问题准确率69%。两项基准共同表明，AI已在部分最困难的生物学问题上超越人类专家。

AK@_akhaliq · 4月30日39

OmniShotCut Holistic Relational Shot Boundary Detection with Shot-Query Transformer paper: https://huggingface.co/papers/2604.24762

译OmniShotCut 基于Shot-Query Transformer的整体关系性镜头边界检测 论文: https://huggingface.co/papers/2604.24762

Anthropic@AnthropicAI · 4月30日51

New on the Science Blog: We gave Claude 99 problems analyzing real biological data and compared its performance against an expert panel. On 23 problems, the experts were stumped. Our most recent models solved roughly 30% of those—and most of the rest.

译科学博客上新：我们给Claude出了99个分析真实生物学数据的难题，并将其表现与专家小组进行了比较。 在23个问题上，专家们被难住了。我们最新的模型解决了其中大约30%——以及其余的大部分问题。