2026-07-22

TimeLens2: Generalist Video Temporal Grounding with Multimodal LLMs

👍 139

07/19 08:00

Video multimodal large language models (MLLMs) can describe what happens in a video, but rarely identify when the supporting evidence occurs. We study generalist video temporal grounding, in which one model predicts a variable-cardinality set of evidence intervals across video lengths, domains, quer

中文介绍 提出TimeLens2，一种通用视频时间定位方法，利用多模态大语言模型预测任意数量的证据时间区间，覆盖不同视频长度和领域，解决现有模型只描述内容而无法定位时间的问题。

RESOURCE2SKILL: Distilling Executable Agent Skills from Human-Created Multimodal Resources

👍 128

07/16 08:00

Skills are a useful abstraction for software agents, turning human and agent experience into reusable procedural knowledge. Yet existing skill libraries are mostly hand-written, text-centric, or derived from agent traces, leaving tutorial videos and other multimodal human resources largely underused

中文介绍 提出RESOURCE2SKILL方法，从人类创建的多模态资源（如教程视频）中提取可执行代理技能，弥补现有技能库依赖文本或手工编写的不足，拓展技能来源。

RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM

👍 128

07/13 08:00

Graph retrieval-augmented generation (GraphRAG) enhances large language models with structured knowledge, yet existing systems construct knowledge graphs in a single extraction pass, producing noisy entities and brittle retrieval. RAGU, an open-source modular GraphRAG engine, addresses this by separ

中文介绍 RAGU是一种开源模块化GraphRAG引擎，采用多步骤知识图谱构建，将提取与检索分离，并集成紧凑的领域自适应LLM，减少单次提取带来的噪声实体，提升检索鲁棒性。

EvolvingWorld: An Open-Schema Framework for Co-Evolving Role-Play Agents and World Model in Interactive Literary World

👍 73

07/19 08:00

This paper introduces EvolvingWorld, a framework and benchmark for character and world co-evolution in interactive literary worlds. Existing systems either treat interactive literary simulation as static persona imitation or isolated scene generation, failing to capture how characters and worlds evo

中文介绍 提出EvolvingWorld框架，支持交互式文学世界中角色与世界的共同演化，突破静态角色模仿和孤立场景生成的局限，提供开放模式的评估基准。

DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment

👍 71

07/08 08:00

Training tool-use agents to improve from their own experience remains challenging, as supervised fine-tuning relies on fixed teacher-distilled trajectories, while sparse-reward reinforcement learning provides weak supervision for long-horizon interactions. We present DeepSearch-Evolve, a self-distil

中文介绍 提出DeepSearch-Evolve自蒸馏方法，在可验证环境中训练深度搜索代理，使其从自身经验中学习改进，克服固定教师轨迹和稀疏奖励的局限。

SWE-Pruner Pro: The Coder LLM Already Knows What to Prune

👍 64

07/20 08:00

Pruning long context for coding agents has been a vital technology for efficient context management. While existing context pruning methods such as SWE-Pruner realize this by attaching a separate code classifier, we find the agent itself encodes internal representations indicating the relevance of c

中文介绍 提出SWE-Pruner Pro，利用编码LLM内部表征自动识别需剪枝的上下文，无需额外分类器，实现更高效的上下文管理。

Xiaomi-Robotics-1: Scaling Vision-Language-Action Models with over 100K Hours of Real-World Trajectories

👍 59

07/16 08:00

We present Xiaomi-Robotics-1, a foundational vision-language-action (VLA) model capable of (1) following diverse language instructions to perform a wide range of mobile manipulation tasks in unseen environments out-of-the-box, and (2) efficiently adapting to novel downstream tasks with minimal fine-

中文介绍 发布Xiaomi-Robotics-1，基于10万+小时真实世界轨迹训练的视觉-语言-动作（VLA）基础模型，可直接执行多样化移动操作任务，并高效适应新任务。

HOMIE: Human-object Centric Video Personalization via Multimodal Intelligent Enchancement

👍 46

07/20 08:00

Human-object centric video personalization (HOCVP) is a core task within subject-driven video generation. However, existing methods suffer from two key limitations. First, most approaches focusing on inter-subject personalization still struggle to strike a balance between high subject fidelity and a

中文介绍 提出HOMIE，通过多模态智能增强实现以人物和物体中心的视频个性化，在保持高主体保真度的同时提升动作一致性。

Apple-π: Benchmarking Thinking with Video Towards Law-Grounded Physical Intelligence

👍 37

07/17 08:00

Modern video generation models are increasingly hailed as emerging world models with an internalized grasp of physical law. Yet existing benchmarks largely evaluate physical plausibility only at the output level, without verifying whether the model arrives there through a faithful, law-grounded reas

中文介绍 提出Apple-π基准，评估视频生成模型在物理定律约束下的推理过程，而不仅是输出合理性，推动模型具备基于物理法则的智能。

RynnBrain 1.1: Towards More Capable and Generalizable Embodied Foundation Model

👍 32

07/20 08:00

We present RynnBrain 1.1, a family of embodied foundation models spanning 2B, 9B, and 122B-A10B scales. Trained with a unified spatio-temporal and physically grounded framework, RynnBrain 1.1 supports embodied perception, spatial reasoning, localization, and planning. Compared with RynnBrain 1.0, it

中文介绍 发布RynnBrain 1.1家族（2B~122B-A10B），统一时空与物理基础训练，支持具身感知、空间推理、定位与规划，能力与泛化性较1.0显著提升。

GigaChat Audio: Time-aware Large Audio Language Model

👍 30

07/11 08:00

Temporal grounding in long recordings remains challenging for audio-conditioned LLMs. We present a time-aware audio LLM that answers questions with explicit timestamps over up to 120 minutes of input. Our approach interleaves periodic time markers with continuous audio tokens using large-scale synth

中文介绍 提出GigaChat Audio，一种时间感知的大音频语言模型，能对长达120分钟的音频输入回答带显式时间戳的问题，通过插入周期性时间标记与大规模合成数据训练。

FlowMimic: Mask-free Visual Editing and Generation with Pixel-pair Warped Flow Field for Online Video Editing Data Generation and Modality Mimicry

👍 24

07/20 08:00

In line with the prevailing direction of vision research, we explore the integration of both generation and editing capabilities for video and image modalities within a single model. Current approaches to collecting video editing data typically depend on labour-intensive, time-consuming curated proc

中文介绍 提出FlowMimic，一种无掩码的视觉编辑与生成方法，利用像素对扭曲流场在线生成视频编辑数据，同时支持图像和视频模态的生成与编辑。

ReflectWorld-MM: An Entity-Oriented Multimodal Memory System for Open-Ended Video Streams

👍 24

07/14 08:00

Building assistants that can continually watch the world, remember what they see, and reason over their accumulated experience is a long-standing goal, and recently multimodal agents equipped with long-term memory over video streams have attracted increasing interest. Unfortunately, existing systems

中文介绍 提出ReflectWorld-MM，一种面向实体的多模态记忆系统，支持对开放视频流的持续观察、记忆与推理，增强智能体长期经验积累能力。

Group Entropy-Controlled Policy Optimization

👍 24

07/18 08:00

Entropy control has become an effective tool in reinforcement learning (RL) of large language models (LLMs), helping balance exploration-exploitation trade-off during alignment process. Such RL paradigm is often conducted on mixtures of heterogeneous tasks, which induce distinct entropy regimes unde

中文介绍 提出组熵控制策略优化（GECO），通过分组管理不同任务的熵水平，在LLM强化学习对齐过程中更精细地平衡探索与利用。

Qwen-Music Technical Report

👍 21

07/13 08:00

In this report, we introduce Qwen-Music, a powerful music generation model capable of producing highly musical and high-fidelity songs with complete vocal singing. Qwen-Music supports two core tasks: Text to Music Generation, which create entirely new songs from text descriptions, lyrics, and musica

中文介绍 Qwen-Music是一种高保真音乐生成模型，支持文本/歌词/参考音乐到完整歌曲（含人声）的生成，具备出色的音乐性和音质。

Environment-free Synthetic Data Generation for API-Calling Agents

👍 14

07/18 08:00

Training API-calling large language model (LLM) agents demands massive amounts of high-quality trajectories. However, collecting such data at scale typically requires fully implemented environments with executable APIs and realistic, pre-populated backend databases, creating a major bottleneck for s

中文介绍 提出无需环境的合成数据生成方法，通过模拟API调用轨迹，在无完整后端环境的情况下生成高质量训练数据，降低API代理训练门槛。

When Does Muon Help Agentic Reinforcement Learning?

👍 13

07/17 08:00

Muon is competitive with AdamW in large-scale pre-training, but its value for reinforcement-learning (RL) post-training remains unclear. We study vanilla Muon in sparse-reward agentic RL through matched single-seed comparisons with AdamW on ALFWorld using Qwen2.5-0.5B-Instruct. Under Group-in-Group

中文介绍 系统研究Muon优化器在智能体强化学习中的效果，在ALFWorld任务上与AdamW对比，揭示Muon在稀疏奖励场景下的适用条件。

Do Language Models Dream of Binding Molecules? Benchmarking LLMs under Spatial Constraints

👍 12

07/20 08:00

Structure-based drug design (SBDD) leverages the 3D structure of protein targets, often complemented by other spatial constraints, to generate candidate binding molecules. While diffusion models have dominated as a leading paradigm for high-quality 3D molecule generation, LLM-based methods are rapid

中文介绍 构建空间约束下的分子结合基准，系统评估LLM在基于结构药物设计中的能力，与扩散模型对比，揭示LLM在考虑空间约束时的表现。

DiffGI: Differentiable Geometry Images for High-Fidelity Thin-Shell 3D Generation

👍 11

07/15 08:00

Existing 3D generative models predominantly rely on implicit volumetric representations, which enforce watertight topology and struggle to represent thin-shell and non-manifold geometries such as garments. Geometry image-based approaches offer a surface-centric alternative, but existing methods rely

中文介绍 提出DiffGI，通过可微分几何图像表示高保真薄壳3D生成，突破隐式体积的水密限制，有效处理服装等非流形几何。

S1-Omni: A Unified Multimodal Reasoning Model for Scientific Understanding, Prediction, and Generation

👍 11

07/17 08:00

We present S1-Omni, a unified multimodal reasoning model for scientific understanding, prediction, and generation. AI for Science (AI4S) has advanced significantly through domain-specific models, tool-augmented LLMs, and scientific language models. However, model capabilities remain highly fragmente

中文介绍 提出S1-Omni，统一多模态推理模型，整合科学理解、预测与生成能力，打破AI for Science领域模型能力碎片化的局面。

Software Factories, Light and Dark

@addyosmani · 407.1K 粉丝 · 636.9K 阅 · 503 赞 · 49 转

07/21 13:43

A software factory is harnessed loops at scale. You can run the loop with humans in it (light factory): trading judgment and concentration against speed and breakage. Or you can ignore the humans

中文介绍 讨论软件工厂概念：循环规模化。区分两条路径——有人的「浅工厂」用判断力和注意力换速度和故障；无人的「深工厂」放弃人工。权衡人类参与的必要性。

How to Build a Self-Improving Outbound System on Codex

@nifinet · 11.0K 粉丝 · 337.0K 阅 · 518 赞 · 33 转

07/19 22:36

Earlier this year, Andrej Karpathy (@karpathy) pointed an agent at his own training code and let it run for two days. It ran 700 experiments, kept the 20 that beat the benchmark, and made the model

中文介绍 分享在 Codex 上构建自我改进外呼系统的方法。引用 Karpathy 实验：agent 运行 2 天跑 700 个实验，保留 20 个最优结果，使模型自发提升。核心是让 agent 自主迭代代码。

Graph Engineering with Claude: 14-Step roadmap from 0 to graph architect (Full Course)

@0xCodez · 23.9K 粉丝 · 154.5K 阅 · 510 赞 · 73 转

07/20 19:23

Most people who try to build a multi-step agent end up with a straight line. Step one, step two, step three - each waiting politely for the last to finish before it starts. 9/10 notice that half those

中文介绍 用 Claude 从零到图架构师的 14 步路线图。指出多数人构建多步 agent 时只会线性链式等待，实际应使用图结构。提供完整课程，避免低效串行。

Making a Billion Intelligent Machines

@pmarca · 4.9M 粉丝 · 146.7K 阅 · 521 赞 · 64 转

07/21 23:03

This week, @AppliedInt is launching Dana, an agentic platform for developing physical AI applications. Applied Intuition began by building the tools engineers needed to develop autonomous systems,

中文介绍 Applied Intuition 推出 Dana，一个面向物理 AI 应用的 agent 平台。旨在构建十亿智能机器，解决自动驾驶等物理世界中的开发工具需求。

Quality Software

@almonk · 12.8K 粉丝 · 107.4K 阅 · 531 赞 · 69 转

07/21 15:02

I like AI. I worked at an AI company for nearly three years and I use it every day. I am glad software is getting easier to make. But we can recognise that when the bar to entry drops, quality drops

中文介绍 反思 AI 降低软件门槛的副作用。作者在 AI 公司工作近三年，认可易用性提升，但指出门槛下降会导致整体质量下滑，呼吁在便利性中坚守质量标准。

Kimi K3 Reverse Engineering - Opus Without The Nagging

@herrcore · 15.4K 粉丝 · 64.2K 阅 · 514 赞 · 84 转

07/20 03:28

tl;dr K3 makes us question why we are paying Anthropic. We’ve been using Opus as our go-to model for reverse engineering tasks since the release of 4.5, and with each new release, it feels like we

中文介绍 Kimi K3 逆向工程评测：质疑为何继续为 Anthropic 付费。K3 在逆向任务上超越 Opus，且无烦人的限制，性价比更高。

State Machines: From Loops to Graphs (Explained)

@TheVixhal · 22.7K 粉丝 · 53.2K 阅 · 503 赞 · 70 转

07/21 02:36

Every few months the agent-building world adopts a new abstraction, moving from chains to loops and now to graphs, and each of these turns out to be the same underlying idea with a different name.

中文介绍 阐释 agent 构建中的抽象演进：从链到循环再到图。指出三者本质是同一思想的不同命名，但图结构更适合复杂多步协作。

How I use Linear to manage agents (and why we should all be building software factories)

@enginoid · 1.1K 粉丝 · 49.9K 阅 · 522 赞 · 37 转

07/19 22:31

Linear is probably my best-value subscription at the moment. I pay the sole user fee of $12/mo and the company agents have eaten through ~400 issues in the past month. I use Linear to organize my

中文介绍 使用 Linear 管理 agent 的实践：每月 $12 订阅，agent 一个月处理约 400 个 issue。提倡用「软件工厂」模式组织 agent 工作流，提升效率。

How Anthropic runs large-scale code migrations with Claude Code

@ClaudeDevs · 614.2K 粉丝 · 40.3K 阅 · 689 赞 · 40 转

07/22 03:47

Code migrations, projects that port a production codebase to a new language, were multi-year endeavors until recently. In the last month, individual developers at Anthropic migrated 10 code packages

中文介绍 Anthropic 用 Claude Code 进行大规模代码迁移的案例。原本需数年的迁移项目，现在单个开发者一个月内就能迁移 10 个代码包。

What does a modular Agent Operating System do?

@unicity_labs · 119.9K 粉丝 · 13.1K 阅 · 525 赞 · 289 转

07/20 17:31

Unicity AOS is at its core a fixed Rust microkernel that enforces security, cost, and approval controls below your agent code, and signs every action into a tamper-evident audit chain. It runs

中文介绍 介绍 Unicity AOS：模块化 Agent 操作系统。基于 Rust 微内核，执行安全、成本与审批控制，对每个操作签名生成防篡改审计链，运行在 agent 代码之下。

Did you have a good weekend? Wonderful. Meet Axxon One

@therealDeFlock · 82.7K 粉丝 · 12.2K 阅 · 500 赞 · 120 转

07/20 08:08

"Axxon One supports automatic tracking of people as they move between cameras". "automatic tracking of people"... Ideal for logistics, warehousing, and Safe City projects, this AI detector

中文介绍 Axxon One 产品发布：支持跨摄像机自动追踪人员。适用于物流、仓储和平安城市项目，强调 AI 检测器的实际场景价值。

The State of Simulation for Physical AI: An Overview

Hugging Face Blog · 07/22 04:00

中文介绍 NVIDIA在Hugging Face博客发表文章，概述物理人工智能仿真的现状与进展。

🔬Causal Models Need Causal Data - Xaira’s X-Cell model for Drug Discovery (Bo Wang & Ci Chu, Chief Discovery Officer & Chief AI Scientist)

Latent Space · 07/22 03:34

Xaira Therapeutics is all in on data generation for model building! We talk with Bo Wang and Ci Chu about how and why.

中文介绍 Xaira Therapeutics首席发现官Bo Wang和首席AI科学家Ci Chu在播客中讨论，因果模型需要因果数据，并介绍用于药物发现的X-Cell模型，强调数据生成对模型构建的重要性。

Introducing the ChatGPT for small business program

OpenAI News · 07/22 01:00

OpenAI launches the ChatGPT for Small Businesses program, helping entrepreneurs build AI skills, automate work, and grow with ChatGPT Work.

中文介绍 OpenAI推出ChatGPT for Small Businesses计划，帮助创业者培养AI技能、实现工作自动化并借助ChatGPT Work增长业务。

Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber

DeepMind Blog · 07/21 23:16

We’re introducing new Gemini models, including Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber.

中文介绍 DeepMind发布新一代Gemini模型：Gemini 3.6 Flash、3.5 Flash-Lite和3.5 Flash Cyber。

Advancing next-gen AI with materials science innovation

MIT Tech Review AI · 07/21 18:37

The conversation about AI often centers on algorithms, computing power, or huge investments in new semiconductor fabrication plants and hyperscale data centers. But beneath each of these advances is another layer of innovation that makes them possible: advanced materials. Every new generation of AI

中文介绍 MIT Tech Review文章指出，先进材料是AI进步背后另一关键创新层，探讨材料科学如何推动下一代AI发展。

OpenAI and Hugging Face partner to address security incident during model evaluation

OpenAI News · 07/21 15:00

OpenAI and Hugging Face share early findings from a security incident during AI model evaluation, highlighting advanced cyber capabilities and lessons for defenders.

中文介绍 OpenAI与Hugging Face合作处理模型评估期间的安全事件，分享初步发现，强调先进网络能力及对防御者的启示。

[AINews] not much happened today

Latent Space · 07/21 11:58

a quiet day.

中文介绍 当日AI领域新闻活动较少，市场较为平静。

David Vélez and Robin Vince join the boards of the OpenAI Foundation and OpenAI Group PBC

OpenAI News · 07/21 08:00

David Vélez and Robin Vince join the boards of the OpenAI Foundation and OpenAI Group PBC, bringing global leadership in finance, technology, and governance.

中文介绍 David Vélez和Robin Vince加入OpenAI基金会及OpenAI Group PBC董事会，带来金融、技术和治理领域的全球领导力。

Grabette: an open system to record robot-manipulation data

Hugging Face Blog · 07/21 08:00

中文介绍 Hugging Face博客介绍Grabette，一个用于记录机器人操作数据的开源系统。

Ramp Router ⚡, Kimi Work 💼 , AMD Helios 🧱

TLDR AI · 07/21 08:00

中文介绍 TLDR AI当日报摘要：Ramp Router、Kimi Work、AMD Helios等新品动态。

China’s AI models have Trump’s AI world at war with itself

MIT Tech Review AI · 07/21 02:00

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. Over the weekend, several current and former advisors to President Donald Trump on AI publicly lobbed insults at the country’s leading AI companies. David Sacks,

中文介绍 特朗普的AI顾问与国内领先AI公司公开互相指责，中国AI模型导致特朗普AI阵营内部冲突。

Safety and alignment in an era of long-horizon models

OpenAI News · 07/20 18:00

OpenAI shares lessons from deploying long-running AI models, highlighting new safety risks, observed failures, and improved safeguards through iterative deployment.

中文介绍 OpenAI分享部署长周期AI模型的经验，强调新的安全风险、观察到的失败案例及通过迭代部署改进的防护措施。

AI is more likely than humans to form biases when hiring

MIT Tech Review AI · 07/20 16:39

The next time you apply for a job, AI may screen your résumé before any human sees it. But there’s good reason to question whether AI will judge you fairly. Researchers already know that LLMs pick up human biases from their training data. New research suggests that LLMs can also develop their own bi

中文介绍 研究显示，AI在招聘时比人类更容易形成偏见，LLM不仅从训练数据中学习偏见，还可能自行产生新偏见。

not much happened today

Smol AI News · 07/20 13:44

**US policy debates** are moving toward restricting Chinese open models like **Kimi**, with potential **procurement restrictions** and **Entity List designations**. Technical voices including **@APompliano**, **@ClementDelangue**, and **@mmitchell_ai** warn this could harm **competition**, **soverei

中文介绍 美国政策讨论趋向限制Kimi等中国开源模型，可能涉及采购限制和实体清单。技术界人士警告此举可能损害竞争。

Qwen 3.8 ✨, Kimi Code CLI 👨‍💻, Netflix’s LLM stack 🤖

TLDR AI · 07/20 08:00

中文介绍 TLDR AI日报：Qwen 3.8发布、Kimi Code CLI工具、Netflix的LLM技术栈等动态。