2026-09-07

Terminal-Universe: Turning Agent Trajectories into Scalable Terminal Environments

👍 272

09/03 08:00

As terminal-based code agents become prevalent, agent trajectories have accumulated at scale, while realistic, executable environments remain scarce. However, environments are what agent post-training actually requires: each can be re-queried into many verifiable tasks and provides execution feedbac

中文介绍 针对终端智能体轨迹丰富但真实可执行环境稀缺的问题，提出Terminal-Universe，将轨迹转化为可扩展的终端环境。每个环境可被重新查询生成多个可验证任务并提供执行反馈，为智能体后训练提供规模化数据源。

Random Attention: Rethinking KV Cache Eviction for Efficient Reasoning

👍 163

09/03 08:00

Large language models achieve superior performance on tasks that require extended reasoning, but long chains of thought make the KV cache a severe memory bottleneck. Existing KV cache compression methods share one paradigm: score each cached token by some estimate of how much it will matter later, a

中文介绍 大模型长链推理使KV缓存成为内存瓶颈。现有压缩方法基于未来重要性打分，但推理中重要性动态变化。提出随机注意力方法，重新思考缓存淘汰策略，通过随机采样替代固定评分，有效降低内存占用，同时保持推理性能。

Knowing When Not to Reuse: Conditional Experience Transfer in Autonomous LLM Post-Training

👍 150

08/27 08:00

Large language models offer broad capabilities, but adapting them to evolving domains, tools, and requirements often entails repeated post-training. Autonomous systems automate parts of this process by proposing updates, training candidates, and using evaluation feedback to select subsequent proposa

中文介绍 针对LLM后训练中重复适应新领域成本高的问题，提出条件式经验迁移方法。在自主后训练流程中，根据当前领域与历史经验的相似度，决定是否复用之前学到的优化策略，减少不必要的重复训练，提升适应效率。

Rethinking On-Policy Distillation of Large Language Models II: One Training Example

👍 78

09/03 08:00

On-policy distillation (OPD) combines student-generated rollouts with dense token-level supervision from a teacher. Existing work has mainly studied its algorithmic behavior, leaving the role of training data unclear. We examine this role at the data-minimal limit by training on a single query. One-

中文介绍 研究在线蒸馏中训练数据的作用，在极简数据极限下仅用单个查询训练。发现单样本训练可揭示OPD算法的核心行为特性，有助于理解样本数量与蒸馏性能的关系，为数据高效蒸馏提供新视角。

Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM

👍 74

09/03 08:00

Hybrid LLMs pair softmax attention with linear-attention layers such as Gated DeltaNet (GDN), whose recurrent state summarizes the context in fixed size. Early community 4-bit quantizations of Qwen3.8-27B (48 GDN layers, 16 attention layers) left the GDN block in 8- or 16-bit precision -- especially

中文介绍 研究混合LLM中线性注意力层Gated DeltaNet的4比特量化可行性。早期量化保留其高位精度，发现GDN在NVFP4 W4A4下仍能保持性能，原因在于其门控机制和循环状态的结构特性，从而可在混合27B模型上实现更高压缩比。

Scal3R: Learning Efficient Multi-Relative Pose Query for Scalable Online 3D Reconstruction

👍 46

09/03 08:00

Online 3D reconstruction models perform poorly on long videos. This happens because regressing poses relative to a fixed first-frame anchor forces extrapolation far beyond the training distribution. Small drifts accumulate and amplify into significant geometric collapse. However, we observe that per

中文介绍 在线3D重建在长视频上表现差，因固定首帧锚点的姿态回归导致外推和漂移累积。提出Scal3R，采用多相对位姿查询，随重建进行自适应更新锚点，减少误差累积，实现可扩展的在线重建，提升长视频下的鲁棒性。

Editable Visual Design

👍 42

09/03 08:00

While diffusion base models such as GPT-Image-2 and Nano-Banana exhibit remarkable visual expressiveness, their end-to-end generation inherently yields flattened bitmaps with error-prone text, precluding layer-wise post-editing. Conversely, code-based visual generation via Coding Agents provides pre

中文介绍 扩散模型生成扁平位图难以逐层编辑，代码生成虽可编辑但表现力弱。提出可编辑视觉设计框架，结合两者优势，通过代码生成结构化表示，同时利用扩散模型提升视觉丰富度，实现分层可编辑的视觉输出。

WHALE: A Simple Recipe for Joint Harness-Weight Optimization

👍 30

08/31 08:00

Agent performance depends jointly on the model parameters and the executable harness code that manages context and control flow. Optimizing either component in isolation can leave the system bottlenecked by its frozen counterpart: weight updates can change which harness is effective, while harness u

中文介绍 智能体性能依赖模型参数与执行代码的联合优化，单独优化一方会受限于另一方。提出WHALE，一种简单配方，联合优化模型权重和代码框架，突破瓶颈，提升整体任务表现。

Beyond Retrieval: Progressive Latent Memory Evolution for Streaming Video Understanding

👍 29

09/03 08:00

Streaming video understanding requires multimodal large language models (MLLMs) to process continuous visual inputs and respond to user queries under strict causality and bounded memory. Existing approaches typically compress historical observations into an external memory bank and retrieve query-re

中文介绍 流式视频理解需在有限内存下处理连续输入。现有方法压缩历史信息到外部记忆库，但查询相关检索易忽略时序演变。提出渐进式潜在记忆进化机制，随视频流动态更新记忆，超越单纯检索，提升长视频理解能力。

RealSWE: A Compositional Evaluation of Coding Agents under Realistic User Requests

👍 26

08/31 08:00

Coding agents are now commonly evaluated on the SWE-bench family of benchmarks, whose tasks are built from curated GitHub issues: long, structured, and information-rich. Real user requests, however, are typically far shorter and less structured. To characterize this gap, we define a six-category inf

中文介绍 现有编码智能体评测基于结构化GitHub问题，真实用户请求往往短且非结构化。构建RealSWE基准，定义六类信息缺口，模拟现实请求，评估编码智能体在非理想条件下的表现，揭示评测差距。

Beyond Visual Similarity: Entity-Aligned Retrieval for Knowledge-Based Visual Question Answering

👍 26

08/19 08:00

Knowledge-Based Visual Question Answering (KB-VQA) relies on retrieving external information to answer queries involving long-tail entities. However, existing retrieval pipelines predominantly employ CLIP-style dual encoders, which prioritize surface-level visual similarity over entity-level semanti

中文介绍 知识型视觉问答依赖外部知识检索，但现有CLIP双编码器侧重表面视觉相似性，忽略实体级语义。提出实体对齐检索方法，针对长尾实体进行语义匹配，提升KB-VQA的信息检索准确性。

ZipTok3D: High-Fidelity 3D Tokenization with Compact Token Prefixes

👍 25

09/01 08:00

Compact token sequences are essential for efficient 3D generation. However, existing 3D tokenizers typically organize latent representations either over spatial regions or as fixed-size sets of global tokens, both suffering sharp reconstruction degradation when compressed to extremely low token budg

中文介绍 现有3D分词器在极低token预算下重建质量严重下降。提出ZipTok3D，引入紧凑token前缀，结合空间区域与全局特征，在极低token数下保持高保真3D重建，提升生成效率与质量。

DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training

👍 24

09/03 08:00

Reinforcement Learning from Verifiable Rewards works well when a task has a programmatic checker, but most long-horizon agent domains have none. We work in the outcome-blind setting, where ground-truth success signals are not available. Multi-criteria rubrics are a popular way to supply such a rewar

中文介绍 针对长程任务缺乏可验证奖励的问题，提出DRACO，使用动态评分细则进行细粒度信用分配。在无结果监督环境下，根据多标准评分，动态调整子任务权重，有效指导智能体训练，提升长程任务表现。

CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation

👍 24

09/03 08:00

MLLM-based embedding models remain limited in compositional retrieval, often failing to distinguish scenes containing the same concepts but different attribute-object bindings. Yet the same backbone can resolve such distinctions when used as a cross-attentive reranker, motivating us to distill its c

中文介绍 MLLM嵌入模型在组合检索上受限，难以区分不同属性-对象绑定场景。同一骨干网络作为交叉注意力重排序器能区分，提出CORE方法，通过重排序器蒸馏提升嵌入模型的组合推理能力，无需额外数据。

WorldReward: Reward Modeling for Camera-Conditioned World Models

👍 23

09/03 08:00

Camera-conditioned world models generate interactive videos in which commanded actions should induce the expected scene changes while appearance, geometry, and temporal dynamics remain coherent. Existing rewards assess these requirements separately: geometry-based rewards estimate trajectory executi

中文介绍 相机条件世界模型生成互动视频，现有奖励分别评估几何执行与外观一致性，缺乏联合评估。提出WorldReward，统一建模奖励，考虑相机、几何与动力学协同，提升生成视频的连贯性与可控性。

PACE: Towards Surfacing Hidden Conflicts in User Requests

👍 22

09/03 08:00

Personalized assistants should not only comply with user requests but also assess whether those requests are appropriate given the user's current circumstances. However, prior work has primarily focused on accurately executing requests, overlooking the need for assistants to account for context and

中文介绍 个性化助手应审查请求是否适合用户当前情境，而非盲目执行。提出PACE方法，识别并揭示用户请求中的潜在冲突，结合上下文评估请求，提升助手的安全性。

Using Grounded Theory for Agent Behavior Analysis at Scale

👍 18

08/31 08:00

Understanding agent behavior requires methods that scale to thousands of trajectories and surface new patterns in long, often unfamiliar tasks where pre-built classifiers fall short. We propose to bring grounded theory into agent trajectory analysis: a six-decade-old qualitative method from the soci

中文介绍 为分析大量智能体轨迹并发现新模式，引入社会科学中扎根理论方法，进行自上而下的定性分析，适应长而陌生任务，无需预定义分类器，可扩展至数千条轨迹，揭示行为规律。

Environment Evolution for Terminal Agents

👍 17

09/03 08:00

Scaling interactive and verifiable environments is critical for training terminal agents. As frontier models become more capable, environments synthesized from scratch become less challenging and thus provide limited learning signals. Recent co-evolution methods iteratively synthesize environments n

中文介绍 终端智能体训练需要可交互环境，随着模型能力增强，从零合成环境难度降低，学习信号有限。提出环境进化方法，迭代合成更复杂环境，保持挑战性，提供持续学习信号，提升智能体泛化能力。

Principia: Relational Physics Tests for Video Models

👍 16

09/03 08:00

Evaluating physical reasoning in video models is difficult because absolute motion measurements depend on frame rate, object scale, and camera calibration, all of which are often ambiguous or unavailable in generated video. We propose a different approach. When two objects in the same scene obey the

中文介绍 视频模型物理推理评测受帧率等绝对测量影响。提出Principia，构建关系物理测试，利用同一场景中两物体满足相同物理定律的相对关系，避免绝对测量模糊，更准确评估物理推理。

Select, Compress, Reinvest: A Controlled Study of Visual-Token Allocation in Long-Video MLLMs

👍 15

09/03 08:00

Long-video language models cannot look at every frame: an hour sampled once per second is 3,600 images, and a system keeps only a small fixed slice of that pool. Which frames survive that slice is usually treated as a preprocessing detail; we test whether it should be. Published selectors make the c

中文介绍 长视频模型无法处理所有帧，保留固定切片。研究帧选择是否影响性能，对比不同选择器在受控实验中的效果，分析视觉token分配策略，发现关键帧选择对下游任务有显著影响。

Grok Imagine Odyssey Contest Winners

@imagine · 79.2K 粉丝 · 714.5K 阅 · 532 赞 · 86 转

09/05 04:49

Homer had a lyre. You had Grok Imagine. Thousands of you quoted our launch post and built original Odyssey scenes in Grok Imagine. Image, video, and the voices. Three to five minutes, with at least

中文介绍 Grok Imagine“奥德赛”主题创作大赛公布结果。数千名用户引用发布帖，在 Grok Imagine 中构建了原创奥德赛场面的图像、视频和声音作品，要求时长3-5分钟。奖项聚焦于展示该模型多模态创作能力。

Five things I learned from using Astra

@dkundel · 23.9K 粉丝 · 225.2K 阅 · 520 赞 · 44 转

09/05 04:27

Back in February, getting Codex to build LEGO models took a custom skill, complicated instructions, and hours of iteration. And it was still not great. I came away thinking it wasn’t something models

中文介绍 作者反思使用 Astra（Codex 新功能）的体验。今年2月，用 Codex 搭建乐高模型需自定义技能、复杂指令和数小时迭代，效果仍不佳。对比后认为，Astra 让模型在构建类任务上的能力显著提升，改变了工作方式。

CarryGo, here we go!

@CarryGo_AI · 9 粉丝 · 192.8K 阅 · 1.0K 赞 · 2 转

09/04 18:34

Inside CarryGo lies a world model—helping organizations see, simulate, act and learn as a system. At the product level, CarryGo is a world model, built first for business. At the system level, it is

中文介绍 CarryGo 正式发布，号称内置“世界模型”，面向商业场景。产品层面，它是一个能帮助组织“看见、模拟、行动和学习”的系统；系统层面，则定位为企业级世界模型基础设施。旨在让企业像系统一样感知与决策。

Applied AI Doesn't Work

@vasuman · 96.1K 粉丝 · 109.9K 阅 · 517 赞 · 32 转

09/05 06:17

The world has spent a fortune on AI. For most enterprises, nothing changed. I've spoken to over 300 CEOs, CIOs and CFOs at the largest companies on Earth. I learned most companies apply AI onto their

中文介绍 作者通过与300多位 CEO、CIO、CFO 交流发现，企业花巨资部署 AI 却大多毫无变化，原因是多数公司只是把 AI 硬套在现有流程上，而非重构业务逻辑。核心观点：应用层 AI 不奏效，问题出在组织而非技术。

SARVAM AI Interview Experience and Prep - Backend Engineer (On-Device AI) Internship:

@buzzy_bit · 767 粉丝 · 97.5K 阅 · 500 赞 · 20 转

09/05 15:30

Disclaimer: This post is written purely to share my experience and the way I prepared. It is not a leaked question bank, I'm intentionally not reproducing exact interview questions, and I'm not naming

中文介绍 作者分享 SARVAM AI 后端工程师（端侧 AI）实习面试经历与备考方法。明确声明不泄露真题、不点名面试官，仅从准备思路、考察重点和心态调整角度复盘，适合想投端侧 AI 方向实习的候选人参考。

How I make content with Astra

@nickbaumann_ · 20.0K 粉丝 · 74.8K 阅 · 511 赞 · 38 转

09/05 04:37

I've been using Astra in Codex to edit videos we've shipped on the official ChatGPT Instagram. I've also been using it to pull clips from longer videos and create product animations. And what's really

中文介绍 作者介绍如何在 Codex 中用 Astra 制作内容：用它编辑 ChatGPT 官方 Instagram 已发布的视频、从长视频中抽取片段、制作产品动画。特别强调 Astra 对视频素材的抽象理解能力，让剪辑与创意生成大幅提效。

AI Engineering Skills Map: Using coding agents

@AndrewYNg · 1.9M 粉丝 · 48.9K 阅 · 918 赞 · 136 转

09/04 23:02

A key AI engineering skill is using coding agents. Your skill at steering them both to write code and to carry out non-code tasks, such as analyzing data or managing system operations, allows you to

中文介绍 吴恩达强调 AI 工程师的关键技能是熟练使用编码代理（coding agents）。不仅要会驱动它们写代码，还要让它们执行数据分析、系统运维等非编码任务。掌握“引导”艺术，能极大扩展个人的工程产出边界。

Rethinking skills and prompts for GPT-6 Astra

@pvncher · 35.4K 粉丝 · 44.6K 阅 · 508 赞 · 37 转

09/05 05:44

Coding agents have come a long way, and best practices are changing fast. What used to require a lot of handholding and scaffolding no longer does. If you’ve been using agents in your projects over

中文介绍 作者指出编码代理演进迅速，最佳实践正在剧烈变化：过去需要大量“手把手”脚手架和提示词编排的场景，如今已大幅简化。若长期用代理做项目，应重新审视技能与提示词设计，适配 GPT-6 Astra 时代的新范式。

An Alien Mind

OpenAI News · 09/06 17:00

Jakub Pachocki reflects on increasingly capable AI and the challenge of keeping it aligned. He calls for stronger safeguards and international coordination.

中文介绍 OpenAI新闻发表文章，Jakub Pachocki反思日益强大的AI及其对齐挑战，呼吁加强安全措施和国际协调。

Research acceleration: The view inside OpenAI

OpenAI News · 09/06 16:00

Inside OpenAI, coding agents are reshaping AI research. Explore early data on agent usage, experiment velocity, task complexity, and research acceleration.

中文介绍 OpenAI内部数据显示，编码智能体正重塑AI研究，提升了实验速度与任务复杂度，推动研究加速。

OpenClaw Power, MacBook Simplicity: Five Days With Grok Bot

Latent Space · 09/05 23:01

SpaceXAI’s Grok Bot has the same level of programming power as OpenClaw, but it’s programmable at a different level of abstraction.

中文介绍 Latent Space评测称，SpaceXAI的Grok Bot编程能力与OpenClaw相当，但抽象层次不同，更易编程，兼具MacBook的简洁性。

Architecting memory and storage in the AI era

MIT Tech Review AI · 09/05 02:39

The era of AI inference has arrived. Imagine a healthcare system analyzing millions of data points in real time to accelerate life-saving medical research, or an intelligent assistant instantly resolving thousands of complex customer needs at once. These real-world breakthroughs rely on advanced inf

中文介绍 MIT技术评论指出，AI推理时代已至，如医疗系统实时分析海量数据，或智能助手处理复杂需求，需重新架构内存和存储。

Data from drones in Ukraine is fueling a new Wild West marketplace

MIT Tech Review AI · 09/04 17:25

Battlefields in Ukraine are littered with the remnants of drones, which are now firmly established as a critical weapon of modern warfare. But behind all that wreckage, there’s a new gold mine for the defense sector. The data drones generate will far outlast the wars in which they are used to fight,

中文介绍 MIT技术评论报道，乌克兰战场无人机数据催生新的“狂野西部”市场，数据价值远超战争本身，成为国防领域的新金矿。

[AINews] GPT-6 Astra: OpenAI’s biggest LLM launch of all time

Latent Space · 09/04 13:18

new SOTA computer use and coding, 2.5x pricier per token, but WAY cheaper per task, less monitorable. overall, a very successful launch of OpenAI’s new frontier model class.

中文介绍 Latent Space报道，OpenAI发布GPT-6 Astra，为史上最大规模LLM发布，计算与编码能力达SOTA，每token贵2.5倍，但每任务成本更低，更难监管。

GPT-6 Astra 🚀, Grok Bot Enterprise 🤖, Runway world model 🌎

TLDR AI · 09/04 08:00

中文介绍 TLDR AI简报聚焦GPT-6 Astra、Grok Bot企业版及Runway世界模型等AI最新动态。

GPT-6 Astra: an automated AI Engineer you can hire for <$6 an hour

Latent Space · 09/04 05:09

We spent 20B+ tokens of GPT-6 Astra to explore everything. Here’s our learnings.

中文介绍 Latent Space探索GPT-6 Astra，称其为自动化AI工程师，时薪低于6美元，并分享消耗超200亿token的经验。

Introducing WeatherNext 3, our most advanced and accurate global weather AI model

DeepMind Blog · 09/03 23:02

中文介绍 DeepMind推出WeatherNext 3，为最先进、最准确的全球天气AI模型。

Daybreak for Frontline Defenders: $1B to protect essential services

OpenAI News · 09/03 21:15

OpenAI introduces Daybreak for Frontline Defenders. A $1 billion commitment expands access to frontier cyber AI, training, and support for essential services.

中文介绍 OpenAI启动“Daybreak for Frontline Defenders”计划，承诺投入10亿美元，扩大前沿网络AI、培训和支持，保护关键服务。

NeoMME: an efficient Multimodal-native and Multilingual Encoder

Hugging Face Blog · 09/03 21:13

中文介绍 Hugging Face发布NeoMME，一种高效的多模态原生多语言编码器。

Legora reviewed 41 documents in minutes with GPT-6 Astra

OpenAI News · 09/03 20:00

Legora used GPT-6 Astra to review 41 documents in minutes, find all four planted errors, and improve performance by nearly 40% in this financial-review workflow.

中文介绍 Legora使用GPT-6 Astra在数分钟内审阅41份文件，找出全部4处植入错误，财务审核工作流性能提升近40%。

Playco cut manual fixes 50% prototyping games with GPT-6 Astra

OpenAI News · 09/03 20:00

Using GPT-6 Astra, Playco built three themed game prototypes from one grey box foundation and reported 50% fewer manual fixes than with the previous model.

中文介绍 Playco采用GPT-6 Astra，从单一灰盒基础构建三款主题游戏原型，手动修复比之前模型减少50%。

GPT-6 Astra: A new generation of intelligence

OpenAI News · 09/03 19:00

Introducing GPT-6 Astra, our most intelligent and aligned model yet, with state-of-the-art capabilities across computer use, coding, cybersecurity, and science.

中文介绍 OpenAI发布GPT-6 Astra，为最智能、最对齐的模型，在计算机使用、编码、网络安全和科学方面达到SOTA。

[AINews] Muse Spark 1.3 matches GPT-5.6-Sol, confirming Meta Superintelligence as the newest Frontier Lab, >90% discount for training

Latent Space · 09/03 12:38

an epic comeback story for Meta

中文介绍 Latent Space报道，Meta的Muse Spark 1.3匹配GPT-5.6-Sol，证实超级智能实验室地位，训练成本折扣超90%，堪称史诗逆袭。