2026-08-02

AskChem: Claim-Centered Infrastructure for Chemistry Literature Synthesis

👍 290

07/30 08:00

Chemistry literature synthesis often requires assembling specific findings scattered across many publications, yet existing literature-search systems primarily return ranked document lists. As a result, scientists and AI agents need to locate relevant information, verify their provenance, and assemb

中文介绍 面向化学文献综述，提出以论断为中心的基础设施 AskChem，将分散在多篇论文中的具体发现按语义主张组织，替代仅返回排序文档列表的检索系统，帮助科学家与 AI agent 定位相关信息、验证来源并组装综述。

Qwen-UI-Agent Technical Report: Toward Next-Generation Real-World Centric Foundation GUI Agents

👍 279

07/30 08:00

GUI agents have the potential to become a general purpose executor over existing digital devices. To advance them toward real-world use, we envision agents that operate reliably on real devices, execute workflows across platforms, combine GUI interaction with CLI execution, complete long-horizon tas

中文介绍 报告 Qwen-UI-Agent，面向真实世界的下一代基础 GUI agent。目标是在真实设备上可靠操作、跨平台执行工作流，并将 GUI 交互与 CLI 执行结合，完成长时程任务，推动 GUI agent 从受控环境走向实际应用。

Metis: Memory Foundation Model

👍 255

07/29 08:00

Recent advances in AI agents have increasingly internalized native capabilities into their underlying foundation models, giving rise to multimodal foundation models and large reasoning models. However, agent memory is still primarily implemented through external modules, leaving the native memory ca

中文介绍 针对 AI agent 记忆仍依赖外部模块的问题，提出记忆基础模型 Metis，将记忆能力内化为基础模型的原生能力，统一建模记忆的存储、更新与利用，为智能体提供端到端可学习的长期记忆机制。

Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering

👍 166

07/30 08:00

Recursive self-improvement (RSI) requires AI systems that improve the process of building AI (i.e., AI4AI); machine learning engineering (MLE) offers a concrete, executable testbed for studying this capability. We introduce OpenMLE, an open full-stack system for RSI research in MLE, spanning verifia

中文介绍 提出 Frontis-MA1，面向机器学习工程的 AI4AI 模型，探索递归自我改进（RSI）。配套发布 OpenMLE，一个覆盖可验证训练与评估流程的开源全栈系统，让 AI 系统在机器学习工程任务中持续改进自身构建 AI 的能力。

PhiZero: A World Model Built Around Physical Language

👍 157

07/30 08:00

We introduce PhiZero, a physical world model built around physical language, a compact discrete representation of world-state transitions. Existing physical world models typically predict future videos directly in pixel space, leaving the underlying world dynamics implicit within high-dimensional vi

中文介绍 提出 PhiZero，围绕物理语言构建的世界模型。物理语言将世界状态转移表示为紧凑离散符号，替代直接在像素空间预测未来视频的做法，使底层物理动态显式化，提升世界模型的可解释性与可控性。

DistillAlign: Coordinating Mode Covering and Mode Seeking in Autoregressive Video Distillation

👍 88

07/29 08:00

Existing autoregressive video distillation methods commonly adopt a Distribution Matching Distillation (DMD)-based multi-stage pipeline. However, they typically decouple the initialization and DMD stages -- which then pursue different target distributions -- and judge the intermediate student mainly

中文介绍 针对自回归视频蒸馏中初始化与 DMD 阶段目标分布不一致的问题，提出 DistillAlign，协调 mode covering 与 mode seeking 两种优化方向，使多阶段蒸馏过程对齐，提高视频蒸馏的训练稳定性与生成质量。

VideoCoCo: Code-as-CoT for Physically-Consistent Video Generation via an Agentic Dual-Engine System

👍 64

07/29 08:00

Text-to-video models have achieved remarkable visual quality, yet they still struggle to generate physically consistent dynamics because the temporal evolution of a scene must be inferred implicitly from a highly compressed text prompt. Existing chain-of-thought approaches introduce intermediate pla

中文介绍 提出 VideoCoCo，用代码作为思维链的中间表示，并通过 agentic 双引擎系统生成物理一致的视频。将场景时间演化的物理规则显式编码为代码，指导视频生成，缓解文本到视频模型仅凭压缩提示隐式推断动态的困难。

Memory Decoder at Scale: A Pretrained, Parametric Long-Term Memory

👍 48

07/30 08:00

Decoder-only language models entangle long-term memory and reasoning in a single parameter set, making it difficult to scale memory capacity independently. Memory Decoder introduces a parametric long-term memory module but only studies it at a relatively small scale. In this work, we present Memory

中文介绍 扩展 Memory Decoder 至更大规模，提出预训练的参数化长期记忆模块，与 decoder-only 语言模型的推理参数解耦，使记忆容量可独立扩展，缓解长期记忆与推理纠缠在同一参数集中的瓶颈。

Beacon: Knowing When and How to Perform Agentic Visual Reasoning

👍 46

07/30 08:00

The fundamental goal of agentic visual reasoning is to improve the success rate of multimodal large language models (MLLMs) on complex tasks, rather than merely equipping them with a sophisticated yet inefficient reasoning paradigm. In this work, we rethink agentic visual reasoning through two key d

中文介绍 提出 Beacon，重新思考 agentic visual reasoning 的两个关键维度：何时需要外部推理以及如何进行。目标是提升多模态大模型在复杂任务上的成功率，避免只引入复杂但低效的推理范式，使视觉推理更高效、更有针对性。

CLBench-V: Evaluating Multimodal Context Learning from Grounding to Knowledge Acquisition

👍 44

07/28 08:00

Real-world tasks often require models to learn from task-specific context rather than relying only on pre-trained knowledge. While recent work has highlighted this capability as context learning, existing evaluations mainly focus on textual contexts. In many practical settings, however, the context

中文介绍 提出 CLBench-V，从 grounding 到知识获取全面评估多模态上下文学习。现有评测多聚焦文本上下文，该基准覆盖多模态场景中模型从任务相关上下文学习而非仅依赖预训练知识的能力，填补评测空白。

BM25 Wins at Scale: A Scaling Study of Retrieval-Augmented Generation Paradigms

👍 43

07/30 08:00

Retrieval-augmented generation (RAG) spans lexical and dense retrieval, graph-based indexing, and agentic search, but these paradigms are usually evaluated on different benchmarks at one corpus size, leaving their accuracy-cost scaling unclear. To bridge this gap, we present a controlled study that

中文介绍 通过受控实验系统比较词法检索、稠密检索、图索引与 agentic search 等 RAG 范式在不同语料规模下的准确率与成本。结果表明 BM25 在扩展至大规模时依然表现突出，为 RAG 范式选择提供 scaling 视角的经验依据。

ACE-Data-0: Human-Centric Ambient Capture as Embodied Data Engine

👍 35

07/30 08:00

Embodied intelligence faces a fundamental data bottleneck. Models must capture how first-person perception, whole-body motion, dexterous manipulation, object state, sound, and touch evolve together as humans pursue goals over time. Existing datasets fragment this experience across viewpoints, modali

中文介绍 提出 ACE-Data-0，以人为中心的环绕式采集方案作为具身数据引擎，同步记录第一人称视觉、全身运动、灵巧操作、物体状态、声音与触觉等多模态信号，解决现有数据集跨视角、跨模态碎片化带来的具身智能数据瓶颈。

Beyond Borrowed Histories: Person-Aligned User Simulation for Interactive Role-Playing Evaluation

👍 30

07/30 08:00

Role-playing agents (RPAs) have become one of the most important consumer applications of large language models. Users engage in multi-turn conversations with RPAs for experiences such as emotional comfort, making reliable evaluation essential for measuring capability, comparing systems, and guiding

中文介绍 针对角色扮演 agent 的多轮交互评测问题，提出 person-aligned 用户模拟方法，生成与目标用户个体特征对齐的虚拟用户，用于更可靠地评估 RPA 的能力、对比系统并指导改进，避免借用他人历史数据的偏差。

RefCaptioner: Multi-Reference Image-Grounded Video Captioning

👍 25

07/30 08:00

Existing video captioning models generate natural descriptions of video content but cannot explicitly ground local visual elements to multiple reference images. We introduce multi-reference image-grounded video captioning, a new task requiring factual video descriptions with phrase-level reference g

中文介绍 提出多参考图像引导的视频描述新任务 RefCaptioner，要求模型生成事实性视频描述，并将局部视觉元素在短语级别关联到多张参考图像，弥补现有视频描述模型无法显式对齐外部参考图的不足。

MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis

👍 23

07/29 08:00

Coding agents have made substantial progress on software engineering tasks that modify existing codebases, including bug fixing and feature implementation. However, constructing a complete program from scratch remains a major challenge: even the frontier models evaluated on ProgramBench fully resolv

中文介绍 提出 MindForge，通过无源码程序合成训练小型语言模型完成全生命周期软件工程。针对从零构建完整程序这一难题，让模型在无需现有代码库的环境中学习设计、实现与验证，提升小模型在 ProgramBench 等任务上的完整解决率。

See2Think: Do Multimodal Models Really Use Intermediate Visual States?

👍 22

07/29 08:00

Multimodal large language models increasingly use sketches, annotations, tools, and intermediate images during reasoning, but it remains unclear whether they truly rely on these visual states. Existing benchmarks are limited both by task collections with narrow coverage or partially text-solvable sa

中文介绍 提出 See2Think，系统检验多模态大模型在推理中是否真正依赖草图、标注、工具和中间图像等中间视觉状态。现有基准覆盖窄且部分可纯文本解决，该工作设计更严格的评测以厘清中间视觉状态的实际作用。

SpatialCLI: Learning to Reason With Spatial Tools, Then Without Them

👍 21

07/30 08:00

Vision-language models (VLMs) are increasingly used in embodied agents to interpret visual inputs, reason about spatial relationships, and make task-level decisions based on that reasoning. However, a fundamental capability mismatch remains: general VLMs can reason about the overall task but often m

中文介绍 提出 SpatialCLI，让视觉语言模型先借助空间工具进行空间推理，再脱离工具完成推理的能力迁移。针对 VLM 在具身 agent 中能理解整体任务但缺乏精确空间关系推理的短板，通过两阶段学习提升其空间能力。

β-OPSD: Deriving with Policy Optimization, Training with Self-Distillation

👍 18

07/30 08:00

On-policy self-distillation (OPSD) is a promising approach to improve reasoning language models, but it remains brittle in practice: making it work reliably often requires substantial engineering effort. We identify a structural source of this difficulty: vanilla OPSD is precisely the β=1 member of

中文介绍 发现朴素 OPSD 是 β=1 的特例后，提出 β-OPSD，将策略优化与自蒸馏纳入统一框架并调节二者权重，缓解在线自蒸馏训练推理语言模型时的脆弱性，提升训练稳定性。

SpecFirst: Behavioral Specification Elicitation as a First-Class Step in Agent-Based Program Synthesis from Scratch

👍 18

07/29 08:00

LLM-based agents excel at software engineering tasks where an existing codebase provides context, but constructing a program from scratch remains fundamentally harder. Recent benchmarks such as ProgramBench quantify this gap: given only natural-language documentation and an execute-only binary as a

中文介绍 提出 SpecFirst，将行为规格获取作为 agent 从零程序合成的一等步骤。针对仅有自然语言文档与可执行二进制的 ProgramBench 等任务，先显式挖掘行为规格，再据此生成代码，缩小从零构建程序的差距。

ShadowDancer: Teaching Video World Models Any Action by Learning Unified Dynamics Representations from a Video and Its Shadow

👍 16

07/30 08:00

We present ShadowDancer, a novel approach to any-action, frame-level control of interactive video world models. The obstacle is representational: existing interfaces either encode an action loosely, leaving how it unfolds for the model to improvise, or encode it exactly through structured signals th

中文介绍 提出 ShadowDancer，从原始视频及其 Shadow 中学习统一动力学表示，支持对交互式视频世界模型进行任意动作的帧级控制。该方法缓解了现有接口在动作编码上“松散”与“过度结构化”之间的表征矛盾。

You already own the most valuable thing in shopping. You just can't reach it.

@fridayresearch_ · 1.9K 粉丝 · 1.6M 阅 · 597 赞 · 451 转

07/31 01:30

Your taste is scattered across forty companies that each own a piece and answer to advertisers. FRIDAY is building the version you own — a personal shopping intelligence that lives on your device and

中文介绍 购物品味数据散落在四十家公司手中，各自为广告主服务。FRIDAY 推出可运行在设备上的个人购物智能，让用户真正拥有自己的品味数据，摆脱被商家分割利用的现状。

MiniMax H3: An Open Model Breaking the Boundaries Between Tasks and Modalities

@MiniMax_AI · 107.4K 粉丝 · 42.3K 阅 · 696 赞 · 103 转

07/31 09:53

Today, we're launching MiniMax H3, a general-purpose multimodal generation model. H3 understands unified context across text, images, video, and audio, generating video with native stereo sound, up to

中文介绍 MiniMax 发布开源通用多模态生成模型 H3，统一理解文本、图像、视频和音频上下文，可生成带原生立体声的视频，具备跨模态理解与生成能力，打破任务与模态之间的界限。

AI News Roundup!

@AITECHio · 455.4K 粉丝 · 32.6K 阅 · 512 赞 · 110 转

07/31 19:58

ChatGPT Can Now Talk While Running Your Computer OpenAI has taken AI agents another step forward. You can now have a natural voice conversation with ChatGPT while it operates your computer in the

中文介绍 OpenAI 将 AI agent 推进到新阶段：ChatGPT 在替你操作系统、执行任务的同时，能与你自然语音对话。用户可随时口头调整指令，无需打断工作流，实现边干活边交流，不再只是单向指令。

Anthropic Are Buying Rare Books, Feeding Them Into AI, Then Destroying Them.

@ActionModelAI · 57.6K 粉丝 · 6.8K 阅 · 521 赞 · 390 转

07/30 21:12

It sounds like the plot of a dystopian film. But according to recently released court documents, internal company communications, and a reported $1.5 billion settlement, it's something that has

中文介绍 听起来像科幻电影，但法庭文件和内部通讯显示，Anthropic 疑似购买珍稀书籍，喂给 AI 后销毁，涉及 15 亿美元和解金。这种行为引发版权与伦理争议，若属实将重创模型训练透明度。

[AINews] not much happened today

Latent Space · 08/01 09:38

apart from DeepSeek V4-Flash 0731, a quiet day.

中文介绍 据Latent Space的AI新闻简报，当日AI领域重大动态不多，除DeepSeek V4-Flash 0731外市场相对平静。

Ten advances in mathematics and theoretical computer science

OpenAI News · 08/01 08:00

OpenAI shares new results on long-standing open problems in mathematics and theoretical computer science, including advances in geometry, cryptography, and complexity.

中文介绍 OpenAI分享了数学与理论计算机科学领域的十项新进展，涉及几何、密码学和复杂性理论中的长期开放问题。

Advancing responsible AI across Europe

OpenAI News · 07/31 23:00

OpenAI shares how its safety, security, transparency, and provenance practices support responsible AI governance in Europe. The work will continue as the EU AI Act advances.

中文介绍 OpenAI介绍其安全、安保、透明度和来源管理实践如何支持欧洲负责任AI治理，并表示相关工作将随欧盟AI法案推进而继续。

Building abundant intelligence

OpenAI News · 07/31 23:00

A full-stack approach to making advanced AI more capable, more affordable, and more widely useful.

中文介绍 OpenAI提出“构建富足智能”的全栈方法，目标是让先进AI更具能力、价格更低、用途更广。

Univé builds an AI-ready workforce

OpenAI News · 07/31 15:00

See how Univé built an AI-ready workforce with ChatGPT Enterprise by combining leadership, responsible governance, and employee-led innovation to transform work at scale.

中文介绍 荷兰保险公司Univé借助ChatGPT Enterprise打造AI-ready员工队伍，结合领导力、负责任治理与员工主导创新，实现大规模工作方式转型。

not much happened today

Smol AI News · 07/31 13:44

**DeepSeek** launched the public-beta of **DeepSeek-V4-Flash API**, boasting a significant post-training performance leap without architecture or size changes, achieving a **Terminal-Bench score of 82.7** and nearing **GPT-5.6 Luna's 51** score at about **60% lower cost per task**. The model feature

中文介绍 DeepSeek发布DeepSeek-V4-Flash API公开测试版，称未改变架构与规模的情况下性能大幅提升，Terminal-Bench得分82.7，逼近GPT-5.6 Luna的51分，每任务成本约低60%。

[AINews] GPT 5.6 price cut by 20%-80%: Cost of GPT 5.4 Intelligence dropped 13x in 4 months due to GPT 5.6 recursive self-optimization

Latent Space · 07/31 12:40

Distillation is all you need!

中文介绍 据Latent Space报道，GPT-5.6价格下调20%至80%，其递归自我优化使GPT-5.4的智能成本四个月下降13倍，并称“蒸馏即一切”。

Disrupting a Criminal Scam Operation

OpenAI News · 07/31 08:00

OpenAI disrupted a Cambodia-based scam operation using ChatGPT to support investment, romance, gambling, and impersonation schemes.

中文介绍 OpenAI宣布捣毁一个位于柬埔寨的诈骗组织，该组织利用ChatGPT实施投资、婚恋、赌博和冒充身份等诈骗活动。

Inkling-Small 🧠, GPT-5.6 price cuts 💸, Gemini Robotics 2 🤖

TLDR AI · 07/31 08:00

中文介绍 TLDR AI本期简报关注Inkling-Small模型、GPT-5.6降价以及Gemini Robotics 2等AI领域最新动态。

GPU Management: Why Idle GPUs Are the New Grounded Aircraft

Hugging Face Blog · 07/30 23:09

中文介绍 Hugging Face博客文章探讨GPU管理问题，将闲置GPU比作停飞的飞机，强调算力资源利用效率的重要性。

Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration

DeepMind Blog · 07/30 23:00

Gemini Robotics ER 2 helps robots reason, collaborate, and solve real-world tasks. It represents a step change in video understanding, tool orchestration, and multi-robot collaboration for robotic applications.

中文介绍 谷歌DeepMind发布Gemini Robotics ER 2，助力机器人推理、协作并解决现实任务，在视频理解、工具编排和多机器人协作方面实现跨越式进展。

Ontologies Are So Back: Why AI Agents Are Reviving the Semantic Web

Latent Space · 07/30 19:17

AI engineers are rediscovering ontologies as a way to keep probabilistic agents inside deterministic boundaries.

中文介绍 Latent Space撰文指出，AI工程师正重新重视本体论，将其作为把概率性智能体约束在确定性边界内的方法，推动语义网复兴。

A fundamental flaw leaves LLMs strikingly vulnerable to attack

MIT Tech Review AI · 07/30 18:15

It is impossible to make large language models fully secure against hacks because of a fundamental flaw in how they work, a team of researchers argue in a paper presented at the International Conference on Machine Learning, a top AI conference, this month. The claim has huge implications for the saf

中文介绍 MIT Tech Review报道，研究团队在本月国际机器学习大会上提出，由于工作原理存在根本缺陷，大型语言模型无法实现完全安全、抵御黑客攻击。

Advancing the price-performance frontier with GPT-5.6

OpenAI News · 07/30 18:00

Explore lower GPT‑5.6 pricing for Luna and Terra—and how OpenAI’s more efficient models help enterprises deploy AI workflows at scale.

中文介绍 OpenAI公布GPT-5.6的Luna和Terra版本降价，称更高效的模型将帮助企业大规模部署AI工作流，推进性价比前沿。

not much happened today

Smol AI News · 07/30 13:44

**OpenAI** aggressively cut prices for **GPT-5.6 Luna** by 80% and **Terra** by 20%, introducing a faster **Sol Fast** tier with up to 2.5× lower latency at double the price, improving agent workflow costs by roughly 10×. The **ARC-AGI-3** debate highlighted that the complete agent system, including

中文介绍 OpenAI大幅下调GPT-5.6价格：Luna降80%，Terra降20%，并推出更快的Sol Fast档位，延迟最多降低2.5倍但价格翻倍，智能体工作流成本约改善10倍；文章还讨论了ARC-AGI-3争议。