2026-09-11

Scaling Automatic Research Agents via World Models

👍 434

08/29 08:00

Automating empirical research is a long-standing direction of AI. Recent automatic research (AutoResearch) agents bring this goal within reach, as modern LLMs show the capability to independently implement solutions and learn from the execution outcomes. Behind these gains, post-training (especially

中文介绍 提出基于World Model的自动化科研智能体扩展框架，利用后训练（尤其强化学习）使LLM智能体独立实现方案并从执行结果中学习，提升经验科学研究的全自动完成能力。

Show-Harness: Just a VLM Agent Can Play Robots

👍 134

09/09 08:00

Foundation vision-language models (VLMs) exhibit broad intelligence about the world, yet translating this intelligence into robot control remains challenging. We present Show-Harness, an Embodied Harness that enables VLMs to "play" robots through a compact semantic interface linking intent to action

中文介绍 提出Show-Harness具身框架，使VLM智能体通过紧凑的语义接口将意图映射为动作，无需传统底层控制器即可直接操控机器人执行物理任务。

NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction

👍 127

09/09 08:00

We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP). Alongside NTP, the model learns through Next Concept Prediction (NCP) to predict discrete concepts that span multiple tokens, introducing an explicit and mo

中文介绍 提出NCP-ArchPreview，一种潜在空间语言模型，在标准NTP之外引入Next Concept Prediction，预测跨多个token的离散概念，显式建模长程语义结构。

AgentGrad: Intervention-guided Prompt Optimization for Multi Agent Systems

👍 89

09/08 08:00

Large language model (LLM)-based multi-agent systems (MAS) achieve strong performance by employing specialized multiple agents, yet their performance depends on the prompt design of each agent. For MAS prompt optimization, textual gradient methods that guide prompt updates using natural-language fee

中文介绍 提出AgentGrad，针对LLM多智能体系统的提示词优化方法，用自然语言反馈作为文本梯度引导各智能体提示更新，提升系统整体协作性能。

Programmable World Model

👍 84

09/09 08:00

Recent video world models generate increasingly realistic and interactive visual experiences, yet lack reliable mechanisms for maintaining persistent world state and enforcing programmable rules over extended interactions. We introduce Programmable World Model, a framework that decouples world-state

中文介绍 提出Programmable World Model框架，将世界状态与视频渲染解耦，在持久世界状态上支持可编程规则，解决长交互中状态一致性与规则约束问题。

SpatialBlock: Enhancing Spatial Intelligence in LVLMs via Synthetic Block-Stacking Problem

👍 46

09/07 08:00

Large Vision-Language Models (LVLMs) have achieved strong performance on diverse visual tasks, yet their ability to reconstruct and reason about the 3D structure of the scene depicted in 2D images -- referred to as spatial intelligence -- remains limited. Existing approaches attempt to address this

中文介绍 提出SpatialBlock方法，通过合成积木堆叠问题增强LVLM的空间智能，解决现有模型从2D图像重构推理3D场景结构能力不足的瓶颈。

T1: Terminal Agent Reinforcement Learning for Long-Horizon Tasks

👍 44

09/10 08:00

Agent usage is shifting toward long-horizon tasks such as coding and scientific discovery, among which terminal tasks are especially important. We introduce T1, a Mixture-of-Experts model of 122B total trained with reinforcement learning, operating a real shell in a cloud sandbox for up to 300+ tool

中文介绍 提出T1，一个122B的MoE模型，通过强化学习训练，在云沙箱中操作真实shell执行长时域终端任务，可完成300+工具调用，面向编码与科学发现场景。

EvoSafeHarness: Evolving Model- and Domain-Specific Harnesses for Securing Agents

👍 35

09/05 08:00

Large Language Model (LLM) agents are turning language into real-world effects, making safety necessary against both indirect prompt injections and direct harmful requests. System-level safety harnesses add an enforcement layer beyond model-level defenses, but existing harnesses are usually designed

中文介绍 提出EvoSafeHarness，可演化生成模型和领域特定安全框架，在系统层面对LLM智能体提供超越模型级防御的防护，抵御间接提示注入和直接有害请求。

WearableQA: A Benchmark for Health Reasoning over Real-World Wearable Data

👍 31

09/04 08:00

Recent advances in wearable sensing enable continuous monitoring of physiological and behavioral signals, yet existing benchmarks rarely evaluate whether AI systems can reason over a real user's longitudinal wearable record. We introduce WearableQA, a benchmark comprising 4,084 10-option multiple-ch

中文介绍 提出WearableQA基准，包含4,084道基于真实用户可穿戴设备纵向数据的十选项多选题，评估AI系统对生理和行为信号的持续健康推理能力。

SWE-Bench Pro Verified: A Reliable Benchmark for Software Engineering Agents

👍 21

09/08 08:00

SWE-Bench Pro has emerged as a standard benchmark for evaluating software engineering agents on challenging repository-level tasks. However, our analysis work show that its evaluation is undermined by two sources of unreliability: reward hacking, enabled by leakage of gold solutions or hidden evalua

中文介绍 发布SWE-Bench Pro Verified，针对现有评测中由金解答泄露和隐藏评估数据泄漏引发的reward hacking问题，构建更可靠的软件工程智能体评测基准。

PARSER: Read in Parallel, Reason in Depth for Long-Context LLM Agents

👍 17

09/06 08:00

Sequential memory agents process long documents by reading chunks one after another while maintaining a compact memory state, coupling document traversal to reasoning depth. This coupling introduces sensitivity to evidence placement and ties inference latency linearly to document length. We introduc

中文介绍 提出PARSER方法，解耦文档遍历与推理深度，通过并行阅读长文档实现深度推理，降低长上下文LLM智能体推理延迟对文档长度的线性依赖。

SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?

👍 16

09/08 08:00

While research on recursive self-improvement (RSI) has predominantly automated model training pipelines, reliable autonomous development demands a missing pillar: post-hoc monitoring and auditing to understand what models learn and ensure safe alignment. Mechanistic interpretability tools are essent

中文介绍 提出SAEScientist-Bench，评估AI智能体能否自主开展SAE可解释性研究，填补自主发展中后验监控与模型理解能力评估的空白。

Scores Alone Do Not Prove Discovery: The Discovery Certification Protocol for Auditing AI Research Agents

👍 16

09/07 08:00

AI research agents combine prior knowledge, public sources, and experimental feedback to produce useful results. The Discovery Certification Protocol (DCP) turns claims about these results into executable recovery and feedback tests. Gate 1 validates useful improvement on sealed evaluation. Gate 2 g

中文介绍 提出Discovery Certification Protocol（DCP），将AI科研智能体的发现声称转化为可执行的恢复与反馈测试，通过密封评测和门控验证确认真实贡献。

X-AuT: Progressive Audio-Encoder Compression for Speech LLMs with Cross-Scale Distillation

👍 12

09/10 08:00

Reducing audio-encoder depth lowers the inference cost of speech large language models, but removing complete blocks perturbs the embeddings consumed by the decoder and can cause deletion and premature end-of-sequence errors. We introduce X-AuT, a progressive framework that selects layer combination

中文介绍 提出X-AuT，渐进式音频编码器压缩框架，通过跨尺度蒸馏选择层组合，在降低语音LLM推理成本的同时避免嵌入扰动导致的删除和过早结束错误。

SyncWorld: Visual Calibration Enables World Models as Zero-Shot Simulators

👍 12

09/08 08:00

World models are increasingly used as policy-in-the-loop imagination environments, where reliable rollouts require fine-grained controllability with respect to low-level robot actions. A key obstacle to scaling such models in robotics is that actions are not a universal language in pixel space: chan

中文介绍 提出SyncWorld，通过视觉校准使世界模型成为零样本仿真器，解决机器人动作在像素空间缺乏通用语言的问题，实现与低层动作精确对齐的可靠rollout。

Φ-Bench: Can Large Language Models Engineer the Infrastructure That Powers Them?

👍 11

09/09 08:00

Large language models (LLMs) have demonstrated remarkable capabilities in reasoning and code generation, raising the prospect that they could assist in developing and optimizing the very infrastructure that powers them. However, existing benchmarks mainly focus on isolated kernels, predefined operat

中文介绍 提出Φ-Bench基准，评估LLM能否工程化开发驱动自身的基础设施，超越仅关注孤立内核和预定义操作符的现有评测，覆盖端到端系统优化能力。

DianShi-RxnDB: A Large-Scale, Fine-Grained Organic Reaction Data Platform Built via a Fully Automated Pipeline for Researchers and AI Agents

👍 9

09/06 08:00

High-quality structured organic reaction data are essential for developing artificial intelligence for chemistry (AI4Chem), yet much of this knowledge remains dispersed across patent text, images, and reaction schemes. We present DianShi-RxnDB, a large-scale, fine-grained organic reaction data platf

中文介绍 提出DianShi-RxnDB，通过全自动化流水线构建大规模细粒度有机反应数据平台，整合分散在专利文本、图像和反应方案中的知识，服务AI4Chem研究。

Memory as Plans: World-Action Modeling with Memory-Grounded Planning

👍 7

09/10 08:00

Mainstream robotic policies often adopt a Markovian formulation, but many complex real-world manipulation tasks are inherently non-Markovian, requiring long-horizon memory beyond the current observation. Existing memory mechanisms often rely on language summaries, growing visual windows, or their co

中文介绍 提出Memory as Plans方法，面向非马尔可夫机器人操控任务，将记忆用于世界-动作建模与规划，避免依赖语言摘要或增长视觉窗口等现有机制的局限。

SchemeArena: Factorized Stress Testing of Scheming in LLM Agents

👍 7

09/08 08:00

We study scheming in LLM agents, in which agents covertly pursue misaligned goals. Our focus is to understand how scheming arises from the interaction of key factors, such as instrumental goals, environmental affordances, oversight conditions, and perceived consequences. Prior work examines only a s

中文介绍 提出SchemeArena，对LLM智能体恶意图谋行为进行因子化压力测试，系统研究工具性目标、环境可及性、监督条件和后果感知等因素如何交互导致图谋行为。

AgenticGen: Reward-Guided Agentic Video Generation for Advertising

👍 7

08/31 08:00

Advertising video generation is not only a video synthesis task, but also a product-conditioned reasoning problem whose success is measured by online business metrics. Recent video foundation models can generate realistic clips from multimodal conditions, yet they do not optimize how a product shoul

中文介绍 提出AgenticGen，奖励引导的智能体式视频生成框架，面向广告场景，将产品条件推理与在线业务指标优化结合，超越纯视频合成的现有生成方法。

10 Ways People Are Making Money With GPT-6 Astra

@sairahul1 · 143.5K 粉丝 · 1.5M 阅 · 561 赞 · 53 转

09/09 18:08

Someone gave GPT-6 Astra $52. 13 hours later: $11,345.95. Someone else gave it $43.47 and one instruction. It turned that into $41,495 overnight. Someone went to sleep. Woke up to $200 in their

中文介绍 博主列举多起用GPT-6 Astra快速盈利的案例：$52在13小时内变成$11,345，$43.47一夜翻成$41,495，另有用户睡醒账户多了$200，聚焦AI自动化套利与被动收入玩法。

How to use ChatGPT Astra for UGC so well it feels illegal

@lucaspatiri_ · 5.8K 粉丝 · 275.5K 阅 · 506 赞 · 41 转

09/09 00:12

openai shipped its most capable model on september 3. i wasn't going to write this. we've generated 2.6 billion organic views across 62 campaigns, managed 590 creators, and produced 22,504 tracked

中文介绍 博主分享用ChatGPT Astra做UGC的完整工作流，已生成26亿次有机播放、覆盖62个campaign、管理590位创作者、追踪22,504个内容数据，强调OpenAI新模型对内容生产的革命性影响。

From GPT-5.6 Sol to GPT-6 Astra: Why Am I Repeating Myself?

@BradGroux · 6.6K 粉丝 · 229.0K 阅 · 561 赞 · 118 转

09/10 06:47

I've been using GPT-6 Astra with the workflow I developed around GPT-5.6: documented task scope, repository context, standard operating procedures, and verification before delivery. In my testing,

中文介绍 博主对比GPT-5.6 Sol与GPT-6 Astra的表现，沿用任务文档、仓库上下文、SOP、交付前验证的工作流，认为新模型优势不如预期明显，质疑自己是否在重复已有流程。

How to design brand mascots with AI (krea). No experience needed

@semochkin_alex · 1.5K 粉丝 · 191.5K 阅 · 507 赞 · 44 转

09/09 20:34

Hi, I'm Alex, cofounder at @relate_studio. In this tutorial, I'll walk you through how we design brand mascots with AI: finding first concepts in Krea, refining the best direction in Nano Banana 2,

中文介绍 Relate Studio联合创始人Alex分享品牌吉祥物AI设计教程：用Krea找初始概念，再用Nano Banana 2精细化方向，全流程无需设计经验，适合品牌方快速产出角色视觉方案。

Pushing Local AI Forward

@MiaAI_lab · 27.4K 粉丝 · 180.6K 阅 · 505 赞 · 58 转

09/09 21:45

For the past three months I’ve been building and shipping local AI recipes, mostly for NVIDIA DGX Spark systems. What started as a personal project has grown far beyond what I expected. My GitHub

中文介绍 博主分享过去三个月为NVIDIA DGX Spark系统构建和发布本地AI配方（recipes）的经历，项目从个人实验发展为GitHub上受欢迎的开源资源，推动本地化AI部署生态。

Reducing cost and improving performance with Claude Platform

@ClaudeDevs · 683.4K 粉丝 · 106.4K 阅 · 1.3K 赞 · 70 转

09/09 01:01

Tuning prompt caching, instructions, and effort can reduce Claude's cost without sacrificing application performance. Performance and cost are often viewed as a trade-off: to spend less, you accept

中文介绍 Claude官方团队分享降本增效策略：通过调优prompt caching、instructions和effort参数，在不牺牲应用性能的前提下降低Claude调用成本，打破性能与成本的二元对立假设。

Personal statement on joining the OpenAI board

@paulfchristiano · 5.3K 粉丝 · 70.7K 阅 · 654 赞 · 96 转

09/10 01:05

I am excited to be joining the OpenAI nonprofit board, serving on the Safety and Security Committee to support safety oversight. Based on the recent trajectory of capabilities and the continued

中文介绍 前Anthropic研究员paulfchristiano宣布加入OpenAI非营利董事会，出任Safety and Security Committee成员，基于近期AI能力轨迹和发展方向，表达参与安全监督的决心。

How to start vibe editing (No Fluff Guide)

@ForwardEditor · 11.7K 粉丝 · 57.6K 阅 · 511 赞 · 32 转

09/11 11:53

Hey, I’m Brian. Let’s get you started with vibe editing. 1. What I highly suggest having DaVinci Resolve Studio - $295 USD. It’s a one-time purchase, not a subscription. In my experience, Studio

中文介绍 Brian发布Vibe Editing入门指南，推荐DaVinci Resolve Studio（$295一次性买断）作为核心工具，强调Studio版优势，提供从零开始用AI辅助视频剪辑的实操路线。

How to Master ChatGPT Images 2.5 (Full Course)

@Av1dlive · 25.6K 粉丝 · 49.2K 阅 · 512 赞 · 53 转

09/10 22:25

i'm going to show you exactly how to master OpenAI's Images 2.5 to create app designs, websites, product shots.....14 tricks. 44 curated examples. the prompts, the results, and how to fix what goes

中文介绍 博主推出ChatGPT Images 2.5完整课程，涵盖14个技巧、44个精选案例，演示如何用该模型生成App设计、网页布局、产品图等，附带prompt模板和常见问题的修复方法。

How to Become a Marketing Engineer

@shannholmberg · 36.1K 粉丝 · 46.0K 阅 · 510 赞 · 44 转

09/10 19:04

marketing engineers build the systems behind the marketing. learn how to connect your knowledge, data and agents into workflows that research, plan and run campaigns. I started bringing more of my

中文介绍 博主分享如何成为营销工程师：将知识、数据和AI Agent连接成自动化工作流，实现调研、策划、投放全链路整合，强调系统化思维而非单点工具能力的重要性。

Software Engineer to AI Engineer: Best Move 2026

@TechWithTimm · 16.2K 粉丝 · 40.2K 阅 · 506 赞 · 79 转

09/09 22:57

A lot of software engineers see AI engineering as a completely different career. But it’s not. I have a friend who works as an AI engineer and makes around $200,000 to $300,000 a year. When I asked

中文介绍 博主分析软件工程师转AI工程师的路径，朋友做AI工程师年薪$200K-$300K，认为两者技能高度重叠，转型门槛比想象中低，属于2026年最具性价比的跳槽方向。

My Son Had Six Weeks to Learn Two Years of Saxon Math. We Used AI.

@DataRepublican · 957.2K 粉丝 · 37.3K 阅 · 846 赞 · 155 转

09/10 10:21

NYC and LA are restricting student AI. We used it to expose what school missed—and repair two years of gaps in six weeks. Part 1 of 3 in a series. Between my husband and I, three of our parents were

中文介绍 博主用AI帮儿子在6周内补完两年的Saxon Math课程，发现学校教育遗漏的知识点并针对性修复，该系列共3篇，探讨AI在个性化教育中的实际效果，与纽约和洛杉矶限制学生使用AI的政策形成对比。

The Complete Guide to pstack Pt. 2

@poteto · 92.6K 粉丝 · 23.3K 阅 · 822 赞 · 60 转

09/10 01:02

In my last post, I showed you why verification is the foundation of everything I do with agents, and how to get started creating your own verification skills. The key lesson was that if an agent can't

中文介绍 poteto发布pstack系列第二部分，延续第一部分的核心理念——验证（verification）是Agent一切工作的基础，本篇深入讲解如何创建自定义验证技能，提升Agent交付可靠性。

The Physical World Does Not Hand You a State Vector

@jhMITgunship · 1.5K 粉丝 · 18.3K 阅 · 1.4K 赞 · 1 转

09/09 01:23

What superintelligence needs to operate in the physical world For much of the work we give models from OpenAI, Anthropic and the other frontier labs, the world has already been turned into context.

中文介绍 作者探讨超级智能进入物理世界面临的根本挑战：现实世界不会给你现成的状态向量（state vector），当前AI处理的多是已结构化的上下文，物理交互需要全新的感知与建模能力。

Alaya AI: Mastering Multimodal Sensor Synchronization for Next-Gen Autonomous Systems

@Alaya_AI · 321.1K 粉丝 · 10.6K 阅 · 504 赞 · 450 转

09/10 18:15

As autonomous vehicles, robotics, and spatial computing platforms evolve, AI model training is shifting away from isolated 2D image annotation toward complex multimodal sensor fusion. Today’s frontier

中文介绍 Alaya AI介绍多模态传感器同步技术，针对自动驾驶、机器人和空间计算平台的需求，AI训练正从单一2D图像标注转向复杂的多传感器融合，面向下一代自主系统的架构演进。

Agents API 🤖, Cognition SWE-2 👨‍💻, Muse Shared Agents 🧑‍🤝‍🧑

TLDR AI · 09/11 08:00

中文介绍 TLDR AI当日汇总：OpenAI Agents API发布、Cognition SWE-2及Muse Shared Agents等动态，无详细内容。

How a researcher uses Codex and ChatGPT to search for new antimicrobial molecules

OpenAI News · 09/11 00:00

César de la Fuente’s lab uses Codex and ChatGPT to search living and extinct genomes for antimicrobial candidates to fight drug-resistant infections.

中文介绍 研究者César de la Fuente团队使用Codex和ChatGPT，在现存及已灭绝生物基因组中搜索抗菌分子候选，以对抗耐药感染。

Now everyone can put data to work

OpenAI News · 09/10 23:00

Meet the Data agent in ChatGPT Work. Connect company data, uncover insights, and build interactive dashboards with AI using natural language.

中文介绍 OpenAI推出ChatGPT Work中的数据Agent，支持连接企业数据，用自然语言挖掘洞察并构建交互式仪表盘。

Powering AI is an architecture problem

MIT Tech Review AI · 09/10 19:00

On July 22, 2026, a transmission line fault in Ashburn, Virginia—the heart of the world’s largest data center cluster—knocked more than 3 gigawatts of load off the grid in seconds. And it wasn’t the first time. Two years earlier, a single failed surge arrester dropped roughly 60 Virginia facilities

中文介绍 MIT科技评论指出AI供电是架构问题：2026年7月弗吉尼亚阿什本一条输电线故障致超3吉瓦负载秒级脱网，此前也曾因单一故障件失效丢约60兆瓦。

Introducing ChatGPT for Financial Services

OpenAI News · 09/10 15:00

Introducing ChatGPT for Financial Services, combining built-in financial data and GPT-6 Astra for research, modeling, and client-ready materials.

中文介绍 OpenAI推出ChatGPT for Financial Services，内置金融数据并搭载GPT-6 Astra，支持研究、建模及生成可直接交付客户的材料。

Expanding AI access and cyber defense for federal, state, local, and tribal governments

OpenAI News · 09/10 15:00

OpenAI and GSA will offer eligible federal, state, local, and tribal governments $0 license fees, 50% off usage, and expanded cyber defense support.

中文介绍 OpenAI与美国通用服务管理局合作，为符合条件的联邦、州、地方及部落政府提供零许可费、五折使用费及扩展网络防御支持。

[AINews] not much happened today

Latent Space · 09/10 11:33

a quiet day

中文介绍 Latent Space AINews当日汇总：当天无重大AI新闻发生。

Build more natural voice experiences with GPT‑Live‑1 in the API

OpenAI News · 09/10 08:00

GPT‑Live‑1 brings natural, full-duplex voice conversations to the API, with stronger instruction following, custom voices, and telephony support.

中文介绍 OpenAI发布GPT-Live-1 API，支持全双工自然语音对话，具备更强指令遵循、自定义音色及电话通信支持。

Introducing the Agents API

OpenAI News · 09/10 08:00

Build and launch cloud agents with the Agents API, a managed service powered by the Codex harness for orchestration, long-running sessions, and tool use.

中文介绍 OpenAI推出Agents API，基于Codex编排框架的托管服务，支持构建云端Agent、长时间运行会话及工具调用。

Rebuilding AUTOMATIC1111 with Gradio Workflow

Hugging Face Blog · 09/10 08:00

中文介绍 Hugging Face博客：使用Gradio Workflow重构AUTOMATIC1111（Stable Diffusion WebUI），无详细技术摘录。

DeepSeek Flash v4.1 🤖, Siri AI 📱, Anthropic AI economics 📈

TLDR AI · 09/10 08:00

中文介绍 TLDR AI当日汇总：DeepSeek Flash v4.1、Siri AI及Anthropic AI经济学等动态，无详细内容。

Paul Christiano joins OpenAI Foundation Board

OpenAI News · 09/10 01:00

Paul Christiano joins the OpenAI Foundation Board and its Safety and Security Committee, bringing experience in AI alignment, safety, and standards.

中文介绍 AI安全专家Paul Christiano加入OpenAI基金会董事会及安全与安保委员会，具备AI对齐与标准制定经验。

IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license

Hugging Face Blog · 09/09 23:36

中文介绍 IBM发布Granite Time Series PatchTST-FM-r2模型，号称时序预测SOTA，采用商业友好许可证。

The AI policy window is open. We need to act.

OpenAI News · 09/09 21:00

Chris Lehane argues that stronger AI capabilities require stronger safety evidence, shared standards, and durable policy action while the policy window remains open.

中文介绍 Chris Lehane发文称AI政策窗口已开启，呼吁以更强安全证据、共享标准和持久政策行动匹配日益增强的AI能力。

GPT-6 Astra: The next generation in intelligence for work

OpenAI News · 09/09 19:00

Meet GPT-6 Astra, OpenAI’s most capable model for business, with advanced reasoning, computer use, and stronger writing and design judgment.

中文介绍 OpenAI发布GPT-6 Astra，定位为面向商业的旗舰模型，具备高级推理、计算机操作能力及更强的写作与设计判断力。