2026-07-27

AREX: Towards a Recursively Self-Improving Agent for Deep Research

👍 139

07/23 08:00

Deep research requires agents to find answers that jointly satisfy multiple constraints. Discovering such answers is costly, whereas verifying a candidate can often be decomposed into tractable constraint-wise checks. This discovery--verification asymmetry suggests that a research agent should do mo

中文介绍 提出AREX，一种递归自改进的深度研究智能体。核心思路是利用发现—验证不对称性，将昂贵的问题发现分解为可验证约束检查，通过递归迭代提升准确性，在复杂多约束问答任务中显著优于现有方法。

SLAI T-Rex: Full-Parameter Post-training of the DeepSeek-V4 Family on Ascend SuperPOD

👍 58

07/22 08:00

Full-parameter post-training of trillion-parameter-scale MoE models introduces substantial system-level challenges for large-scale distributed training, including severe memory pressure, non-overlapped communication overhead, and inefficient kernel execution. While most large-scale LLM training syst

中文介绍 提出SLAI T-Rex系统，在昇腾SuperPOD上实现DeepSeek-V4系列万亿参数MoE模型的全参数后训练。通过消除内存压力、通信重叠和高效内核执行等系统级挑战，显著提升训练效率并保持模型性能。

K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs

👍 57

07/23 08:00

Large language models are increasingly used in K-12 education, but existing benchmarks mainly test exam question answering rather than understanding how curriculum knowledge is structured and visually presented. We call this capability curriculum cognition. It covers prerequisite chains, concept tax

中文介绍 构建K12-KGraph课程对齐知识图谱，用于基准测试和训练教育LLM。该图谱包含先修链、概念分类等课程认知要素，支持结构化知识可视化与评估，填补了现有基准仅测试答题能力的空白。

ReferTrack: Referring Then Tracking for Embodied Visual Tracking

👍 49

07/22 08:00

Embodied visual tracking (EVT) requires a mobile agent to continuously follow a specific target described in natural language using only onboard vision. While recent vision-language-action (VLA) policies unify target identification and trajectory planning, their chain-of-thought (CoT) reasoning ofte

中文介绍 提出ReferTrack，两阶段具身视觉跟踪方法：先通过自然语言定位目标，再使用视觉跟踪模块持续跟随。解耦识别与规划，避免VLA策略中CoT推理导致的错误积累，在真实机器人场景中实现更高跟踪成功率。

Visual Contrastive Self-Distillation

👍 43

07/23 08:00

On-policy self-distillation (OPSD) is promising as it removes the external teacher required by on-policy distillation (OPD), yet it still needs asymmetric information between teacher and student to ensure that the self-teacher provides a stronger learning signal than the student. Existing methods cr

中文介绍 提出视觉对比自蒸馏（VCSD），一种无需外部教师的在线自蒸馏方法。通过对比学习在教师与学生网络间构建非对称信息，使自教师提供更强学习信号，在多个视觉识别任务上达到与有教师蒸馏相当的性能。

Show, Don't Tell: Evaluating Spatial Cognition in Generative Pixels Rather Than LLM Text

👍 35

07/23 08:00

Spatial intelligence is essential for agents to move from static semantic understanding toward interacting with the physical world. Many spatial tasks are grounded in continuous visual scenes, where locations, regions, and paths are more naturally expressed by pointing, marking, or drawing than by r

中文介绍 提出像素级空间认知评估方法，要求模型在生成图像中标注位置、区域和路径，替代文本问答。相比传统LLM文本评估，该方法更贴近真实空间交互，揭示现有视觉-语言模型在连续空间推理上的不足。

Beyond Relevance-Centric Retrieval: Rubric-Oriented Document Set Selection and Ranking

👍 29

07/22 08:00

As large language models and AI agents become the primary consumers of search results, document set quality determines the upper bound of downstream generation. Yet existing evaluation systems remain confined to scoring documents independently and aggregating via nDCG, ignoring inter-document intera

中文介绍 提出基于评分标准的文档集选择与排序方法，超越传统相关性排序。通过考虑文档间交互和整体集合质量，优化多文档下游生成效果，在多个检索增强生成场景中优于基于nDCG的逐文档排序。

NVIDIA-labs OO Agents: Native Python Object-Oriented Agents

👍 26

07/22 08:00

Traditional agent development is split across prompt templates, tool schemas, callback code, and workflow graphs. We present NVIDIA Object-Oriented Agents (NOOA), a model-agnostic Python framework for building reliable AI agents. NOOA takes a simpler approach: an agent is a Python object. Its method

中文介绍 提出NVIDIA OO Agents（NOOA），一种模型无关的Python原生面向对象智能体框架。将智能体视为Python对象，方法为工具，通过原生面向对象编程统一提示模板、工具架构和回调代码，降低开发复杂度。

An Exam for Active Observers

👍 25

07/17 08:00

Human vision is a closed loop: gaze is continuously redirected by intermediate hypotheses rather than a single snapshot. Decades of psychophysics and cognitive science have argued that this active observation is essential for a wide range of tasks. Whether today's multimodal large language models (M

中文介绍 设计主动观察者考试，评估多模态大模型在闭环视觉（如视线重定向）下的表现。通过交互式实验证明，当前LLM在主动观察任务中远逊于人类，揭示其在动态假设驱动视觉处理上的根本缺陷。

LLMs Get Lost in Evolving User Intent

👍 21

07/22 08:00

As LLMs become more capable, they are increasingly deployed as collaborative agents, taking on user-delegated tasks through iterative interaction. Yet genuine interaction is inherently dynamic: users rarely specify their intent upfront, instead disclosing, revising, and reshaping it as the conversat

中文介绍 研究LLM在用户意图动态演化场景中的表现。实验表明，当用户逐步披露、修正意图时，现有模型常迷失方向，无法有效调整行为。提出EvolvingIntent基准，揭示协作交互中的关键鲁棒性瓶颈。

Tencent WorkBuddy Bench: A Multi-Domain Coding-Agent Benchmark with Contamination-Resistant Task Construction

👍 21

07/23 08:00

We introduce Tencent WorkBuddy Bench, a multi-domain evaluation suite for coding agents; this report documents its construction methodology, scoring protocol, and a cross-model leaderboard. At its core is a unified evaluation framework for constructing and running distribution-informed coding-agent

中文介绍 推出Tencent WorkBuddy Bench，多领域编码智能体基准。核心是抗污染的任务构建框架和统一评分协议，覆盖分布内/分布外场景，提供跨模型排行榜，有效评估编码智能体的实际能力。

Self-Supervised Learning of Structured Dynamics from Videos

👍 18

07/23 08:00

Understanding motion in video is a fundamental challenge for visual learning, as frame-to-frame change entangles two sources of dynamics: camera motion and object motion. This decomposition has remained underexplored in representation learning, partly because these factors are tightly coupled in nat

中文介绍 提出自监督视频结构动力学学习框架，从无标签视频中分解相机运动与物体运动。通过解耦表征学习，在多种下游任务（如动作识别、轨迹预测）上显著优于传统自监督方法。

Color Pass-Through via Camera-Display Coupling

👍 18

07/14 08:00

When a real-world scene is captured by a smartphone camera and viewed on its screen, the displayed image often differs noticeably from the original scene in color, brightness, and contrast. This gap persists despite substantial advances in both modern cameras and displays. A key reason is that most

中文介绍 提出摄像头-显示耦合色彩传递方法，校正手机拍摄后屏幕显示的颜色、亮度和对比度差异。通过联合建模相机与显示器的色彩特征，实现更准确的色彩还原，在多种设备组合上验证了有效性。

Scaling Laws for Hypernetwork-Based Knowledge Injection in Large Language Models

👍 16

07/21 08:00

Injecting factual knowledge into large language models (LLMs) reliably and at scale remains an open challenge. Hypernetworks provide a promising solution to large-scale knowledge injection. Although hypernetworks are typically applied for test-time adaptation, we explore their use in train-time know

中文介绍 研究超网络在LLM中大规模知识注入的缩放规律。在训练阶段使用超网络生成参数增量，实现高效知识存储与更新，发现注入性能随超网络容量呈现幂律缩放，为可扩展知识编辑提供理论指导。

Beyond Euclidean Clipping: Overcoming Exploration Collapse in LLM RL via Riemannian Isometric Policy Optimization

👍 13

07/11 08:00

Reinforcement learning (RL) has become a dominant paradigm for enhancing LLMs' reasoning capabilities. However, RL algorithms with PPO-Clip are inherently limited by exploration collapse. Subsequent works remain primarily heuristic and fail to identify the essential cause of PPO-Clip's failure. This

中文介绍 揭示PPO-Clip导致LLM RL中探索崩溃的根本原因，提出黎曼等距策略优化（RIPO）。通过将策略更新限制在概率分布的黎曼流形上，避免欧几里得裁剪引起的探索退化，在数学推理任务中显著提升采样效率和准确率。

Sample-Efficient Learning from Agent Experience

👍 12

07/23 08:00

Real-world agent learning is often constrained by costly environment interactions, such as running time-consuming experiments or obtaining human feedback. In-context learning offers a highly sample-efficient way for agents to learn from their own interaction histories, but its gains disappear once t

中文介绍 提出基于智能体自身交互历史的样本高效学习方法。利用上下文学习快速适应新任务，但指出其增益在上下文变长时消失。设计混合架构，在少量交互后切换为参数更新，显著降低环境交互成本。

Streaming Multi-Agent Autoregressive Diffusion Model with World State Registers

👍 12

07/23 08:00

Multi-agent interactive world models should not only generate consistent observations, but also maintain world states that persist across agents and evolve across views. Existing autoregressive video diffusion pipelines carry forward observation history as conditioning context, which makes shared st

中文介绍 提出流式多智能体自回归扩散模型，引入世界状态寄存器维护跨智能体并随时间演化的共享状态。相比仅拼接观测历史的基线，该方法在多智能体交互视频生成中保持一致性并提升长期预测质量。

Multi-Turn On-Policy Distillation with Prefix Replay

👍 9

07/16 08:00

We study on-policy distillation (OPD) for agentic tasks, where an LLM agent interacts with an environment over multiple turns and a student imitates a teacher over these multi-turn interaction histories. Fully online OPD is costly because each update requires fresh student rollouts through the envir

中文介绍 提出Prefix Replay的多轮在线策略蒸馏方法，用于LLM智能体任务。通过复用前缀轨迹的教师模仿信号，减少在线学生回滚成本，在多个对话和工具使用任务中实现与完全在线蒸馏相当的性能，同时大幅提升训练效率。

Predictive Divergence Masks for LLM RL

👍 9

07/12 08:00

Reinforcement learning for large language models (LLMs) typically relies on trust-region masks to stabilize off-policy updates. The dominant PPO-style approach uses the sampled-token importance ratio for two criteria: a proximity criterion, which asks whether the policy has moved too far from the be

中文介绍 提出预测性差异掩码，为LLM RL中的信任域更新提供更精准的稳定性条件。不仅考虑策略偏移（重要性比值），还引入对优势预测的差异测量，在多个数学和对话任务上比PPO更稳定、更高效。

DocOps: A Verifiable Benchmark for Autonomous Agents in Complex Document Operations

👍 9

07/22 08:00

As autonomous agents rapidly evolve, their ability to reliably manipulate ubiquitous digital documents has become critical for enabling general-purpose AI assistants and automating complex workspace workflows. In this paper, we introduce DocOps, a deterministically verifiable evaluation framework un

中文介绍 引入DocOps，用于复杂文档操作的确定性可验证基准。涵盖多种文档编辑、格式化、提取等任务，提供可重复的评估协议，用于衡量自主智能体在数字文档操作中的可靠性。

Trajectory: A Standard Format for Agent Experience Data

@Letta_AI · 9.8K 粉丝 · 86.5K 阅 · 504 赞 · 31 转

07/24 14:07

Introducing trajectory, an open-source package that normalizes coding-agent sessions from @AnthropicAI Claude Code, @OpenAI Codex, @pidotdev, @LangChain deepagents, @openclaw, Letta Code, and other

中文介绍 Letta_AI发布开源包trajectory，标准化来自Anthropic Claude Code、OpenAI Codex等AI编程代理的会话日志格式，便于分析、复现与比较代理行为，提升工作流可观测性。

Graph Engineering Clearly Explained

@akshay_pachaar · 282.5K 粉丝 · 66.4K 阅 · 515 赞 · 74 转

07/26 02:48

Loop engineering got about six weeks in the spotlight before the timeline moved on. On July 18, Peter Steinberger, the person behind OpenClaw, posted a nine-word question. "Are we still talking loops

中文介绍 讨论AI代理架构趋势：循环工程（loop engineering）热度消退后，行业转向图工程（graph engineering）。引用OpenClaw作者Peter Steinberger的质疑“我们还在谈循环吗？”，指出更结构化的图工程成为新方向。

Why Software Factories Fail

@dexhorthy · 24.8K 粉丝 · 64.9K 阅 · 515 赞 · 50 转

07/25 00:51

or: the harness is not enough Update - the talk version of this post is live on youtube: https://www.youtube.com/watch?v=Ib5GBkD555M i guess we doin loops now We're all racing to put AI coding into production. A lot has

中文介绍 剖析AI编码生产化中“软件工厂”失败的根源，核心观点是仅靠工具框架（harness）不够，需更深入架构设计。附有演讲视频，探讨如何避免常见陷阱。

Agent Harness Engineering vs. Loop Engineering vs. Graph Engineering

@beamnxw · 2.6K 粉丝 · 53.5K 阅 · 561 赞 · 92 转

07/25 22:25

A practical guide to the three architecture layers people keep mixing together The confusion is understandable. All three ideas sit around the same model, all three influence reliability, and all

中文介绍 实用指南厘清Agent Harness工程、循环工程与图工程三个架构层的差异，指出三者围绕同一模型但各有侧重，混淆会导致可靠性问题，帮助开发者正确分层设计。

The new rules of context engineering for Claude 5 models

@trq212 · 318.0K 粉丝 · 48.6K 阅 · 1.2K 赞 · 119 转

07/25 01:45

I’ve written previously about how to best prompt the newest generation of Claude 5 models and work with them iteratively to discover what you want to build. But when you send a message to Claude, the

中文介绍 分享针对Claude 5模型的新版上下文工程（context engineering）规则。强调与模型迭代交互时如何构造对话上下文以提升生成质量，比传统提示工程更关键。

Practical multi-agent orchestration in Codex

@pvncher · 30.4K 粉丝 · 44.0K 阅 · 565 赞 · 33 转

07/25 01:30

GPT-5.6 Sol gets especially interesting when it has a team to work with. Codex's new Multi-Agent V2 tools give Sol and Terra a natural way to delegate tasks, share updates, and coordinate through

中文介绍 演示Codex Multi-Agent V2下多智能体编排：GPT-5.6模型Sol与Terra通过委托任务、共享更新、协调工作流实现复杂任务，展现多智能体协作的实用潜力。

How we teach AI models

@leerob · 272.8K 粉丝 · 32.9K 阅 · 550 赞 · 51 转

07/24 09:39

How do AI models learn new skills and behaviors? The process is surprisingly human and easy to understand, even if you don’t have a machine learning background. Helpful colleagues AI models,

中文介绍 通俗解释AI模型学习技能的过程，类比人类学习。涉及“有帮助的同事”隐喻，可能指向RLHF或微调机制，适合非技术读者理解模型训练原理。

[AINews] Claude Opus 5: Fable-level performance at Opus price (half Fable)

Latent Space · 07/25 15:25

ain't nobody beats Anthropic at distilling Fable!

中文介绍 Anthropic 发布 Claude Opus 5 模型，性能与 Fable 相当但价格仅为一半。

Opus 5

Smol AI News · 07/24 13:44

**Anthropic** launched the **Claude Opus 5** model, which sparked mixed reactions including benchmark scrutiny and praise for its coding-agent capabilities. The model achieved an **Epoch Capabilities Index (ECI) of 159**, slightly below **Fable 5's 161**, but matched Fable 5 on software engineering

[AINews] Black Forest Labs FLUX 3 - Multimodal Flow Models that beat Seedance 2.0, Gemini Omni and Grok Imagine, and FLUX-mimic video-action robotics model

Latent Space · 07/24 12:30

A HUGE win for BFL!

ChatGPT Health 🩺, Fugu-Ultra 1.1 🐡, Runway Media Router 🎬

TLDR AI · 07/24 08:00

How AI helps scientists design the next generation of medicines

MIT Tech Review AI · 07/23 20:00

Designing and developing a new medicine is an expensive, failure-prone scientific challenge. A new drug can take many years to develop, at the cost of a significant investment. And even then, most possible candidates never reach the patient. For biologic medicines, therapies made from engineered pro

not much happened today

Smol AI News · 07/23 13:44

**The Stack v3** is released as the largest open code dataset with **114 TB raw data**, **224M repositories**, and **5T deduplicated tokens**, significantly expanding data for open code models and cyber-defense. The debate on **distillation** continues as a key ideological fault line, with calls for

[AINews] "Laguna S 2.1 Released: Cheaper than Deepseek v4 Flash, Better than V4 Pro"

Latent Space · 07/23 13:18

a quiet day lets us highlight a new neolab win.

Inside the Model Factory — Eiso Kant, Poolside AI

Latent Space · 07/23 13:09

Poolside's co-CEO on how his small team of top researchers built a model factory capable of training Laguna S - a 118B MOE beating Thinky's ~1T open weights model... and this is just the beginning.

Launching Health in ChatGPT

OpenAI News · 07/23 08:00

Health in ChatGPT now lets eligible U.S. users securely connect medical records and Apple Health to get more personalized insights and better understand their health.

Bringing Nunchaku 4-bit Diffusion Inference to Diffusers

Hugging Face Blog · 07/23 08:00

Cursor Router 🔀, OpenAI Presence 🤖, AMD + Anthropic deal 🤝🏻

TLDR AI · 07/23 08:00

Accelerating the frontiers of scientific discovery: Google’s $40M commitment to the Genesis Mission

DeepMind Blog · 07/22 21:38

Google commits $40M in AI tokens and credits for the Genesis Mission

Building AI infrastructure with the Effingham County community

OpenAI News · 07/22 21:00

OpenAI announces Project Camellia in Effingham County, Georgia, with commitments to responsible energy, community investment, jobs, and access to Codex.

How news organizations are using AI to advance their vital missions

OpenAI News · 07/22 21:00

News organizations are using AI to strengthen reporting, grow audiences, and improve business operations, with OpenAI tools supporting journalists and publishers worldwide.

Advancing the next era of national science

OpenAI News · 07/22 20:00

OpenAI outlines its commitment to advancing American science working with the U.S. Department of Energy and national labs to use frontier AI to accelerate discovery.