2026-08-22

EnvHarness: Awakening Static Worlds for Agent Learning

👍 245

08/20 08:00

LLM agents learn by interacting with environments, yet these environments are hand-built and static: blind to an agent's weaknesses, and quickly left behind as it improves. While recent environment generation methods attempt to address this, they require domain-specific pipelines, rely on expensive

中文介绍 EnvHarness 针对 LLM agent 训练中环境静态、无法随能力提升而进化的问题，提出动态唤醒既有世界的环境生成机制，按 agent 弱点自适应调整环境，避免昂贵且依赖领域管线的生成方案，使训练环境持续保持挑战性。

Zetta ζ: An Efficient Closed-Loop Embodied Harness for Self-Evolving Physical Intelligence

👍 140

08/17 08:00

Embodied agents are increasingly used to close the gap left by end-to-end policy models. Yet the agentic path has not realized closed-loop learning in physical execution: existing harnesses remain largely open-loop, following fixed skills during rollout and reflecting only after an episode completes

中文介绍 Zetta ζ 提出面向具身智能体的高效闭环训练框架，解决现有 harness 在物理执行中开环、只能在回合结束后反思的问题，在 rollout 过程中依据执行反馈即时调整技能，支持智能体自我进化式学习，提升物理执行闭环效率。

SemaPLC: A Project-Grounded, Verification-Gated Agent Harness for PLC Code Generation

👍 115

08/19 08:00

Programmable logic controllers (PLCs) run industrial plants, and large language models can already generate independent program organization units (POUs) for them. Whether such logic integrates into an existing PLC project and then runs correctly has been checked only in limited tests. We present Se

中文介绍 SemaPLC 提出面向 PLC 代码生成的项目级 agent 框架，以项目上下文为根基并通过验证门控检查生成逻辑能否集成到既有 PLC 项目并正确运行，弥补此前仅做独立 POU 测试、缺乏集成与执行验证的不足。

FACET: Preserving Source Intent and Executable State in Terminal Task Synthesis

👍 111

08/19 08:00

Training terminal agents requires scalable executable supervision, yet synthesizing high-quality terminal tasks remains challenging. Each task couples an instruction, an initialized environment, a reference solution, and an executable verifier; if these artifacts are generated from inconsistent assu

中文介绍 FACET 面向终端 agent 训练中的任务合成问题，提出保持源意图与可执行状态一致性的生成方法，使指令、初始环境、参考解和可执行验证器来自一致假设，从而规模化提供高质量可执行监督。

Co-RL: Unsupervised Reasoning Emerges from Diverse Cohort in Multi-agent RL

👍 91

08/19 08:00

Reinforcement learning (RL) has emerged as a powerful approach for improving reasoning in language and vision-language models, yet its strongest successes still depend heavily on ground-truth supervision (e.g., verifiable reward). Such annotations are costly to obtain and become increasingly scarce

中文介绍 Co-RL 提出不依赖外部真值奖励的多智能体强化学习推理方法，利用多样化智能体群体构成内在训练信号，使推理能力在群体协作与竞争中以无监督方式涌现，缓解 verifiable reward 等标注稀缺且成本高的问题。

OmniScientist: An Omni-Modal Omni-Discipline AI Scientist

👍 88

08/13 08:00

Recent advances in foundation models have enabled AI scientists to automate increasingly complete research workflows, from hypothesis generation and code execution to manuscript preparation. Yet workflow coverage alone does not provide access to the full evidence on which scientific discovery depend

中文介绍 OmniScientist 提出全模态、全学科的 AI 科学家框架，在自动化假设生成、代码执行和论文写作之外，统一接入并利用多模态、多学科证据，避免因工作流覆盖不足而丢失科学发现所依赖的完整证据链。

SWE-bench Science: Can Coding Agents Resolve Engineering Tasks in Science?

👍 58

08/20 08:00

Software increasingly functions as part of the scientific instrument itself, making failures in scientific code capable of compromising not only program behavior but also the evidence underlying scientific conclusions. Yet existing evaluations of coding agents largely emphasize aggregate task succes

中文介绍 SWE-bench Science 面向科学领域的编码 agent 评估，聚焦科学软件作为仪器组成部分、代码故障可能损害证据可靠性的场景，考察 agent 能否解决真实的科学工程任务，而不只追求总体任务成功率。

SPADE: Self-Play in Adaptive Synthetic Executable Environments

👍 47

08/19 08:00

Continuous self-improvement requires an ever-expanding pool of self-generated, diverse, adaptive goals. For language agents, existing training environment pools (hand-curated, statically synthesized, or frozen-verifier) keep the goal distribution fixed as the learner scales. We introduce SPADE (Self

中文介绍 SPADE 提出基于自博弈的自适应合成可执行环境池，随学习器规模扩大持续生成多样化、自适应目标，打破现有环境池目标分布固定的限制，为语言 agent 的持续自我改进提供动态训练环境。

Training Chemical Plausibility-Aware Large Language Models for Single-Step Retrosynthesis

👍 32

08/19 08:00

Single-step retrosynthesis is a central component of computer-aided synthesis planning, yet its intrinsically one-to-many nature is poorly captured by single-answer evaluation and benchmarking protocols. To address this, we introduce Top-K prompting as a robust training and inference paradigm to bet

中文介绍 针对单步逆合成的一对多本质，提出 Top-K prompting 训练与推理范式，让 LLM 同时建模多个可行逆合成路径并感知化学可行性，替代单答案评估协议，提升逆合成规划的实用性与准确性。

MemTrapBench: Benchmarking Cognitive Traps in LLM Memory Use

👍 31

08/20 08:00

Memory has become a key component of large language models, enabling them to retain information and learn from long-term interactions. However, existing memory benchmarks mainly evaluate whether information is correctly extracted, stored, and retrieved, while largely overlooking how retrieved memori

中文介绍 MemTrapBench 提出评估 LLM 记忆使用中认知陷阱的基准，超越只检验信息提取、存储与检索的传统记忆评测，关注检索到的记忆会否误导推理与决策，揭示记忆在长期交互中的潜在风险。

SkillEvo: Self-Renewing Evolution Gradients from Multi-Turn Interaction Feedback

👍 29

08/13 08:00

Agent Skills are today either hand-authored or produced in a single LLM generation pass, and consequently possess no closed loop through which they might improve from the interaction failures they actually cause. Recent work does close this loop, but derives its feedback from single-turn question-an

中文介绍 SkillEvo 提出从多轮交互反馈中生成自我更新演化梯度的智能体技能框架，让技能能够依据其在实际交互中造成的失败闭环改进，弥补手工编写或单次生成技能缺乏反馈回路、且单轮反馈信息不足的问题。

The More Popular, The Harder to Forget: Adaptive Popularity for LLM Unlearning

👍 17

08/14 08:00

Popular facts are memorised more deeply during pretraining and resist removal longer than rare ones, yet existing LLM unlearning methods apply uniform gradient pressure regardless of training-data frequency. We propose the AdaPop (Adaptive Popularity) method, which combines local token confidence wi

中文介绍 AdaPop 提出自适应流行度的 LLM 遗忘方法，针对高流行事实在预训练中记忆更深、更难移除的现象，根据训练数据频率施加非均匀梯度压力，并结合局部 token 置信度提升遗忘效果，弥补统一力度方法的不足。

Repo0: Design-Driven Zero-to-All Code Generation

👍 16

08/20 08:00

Large language model agents have made substantial progress in code generation, yet most existing systems assume a predefined repository architecture. This assumption does not hold in zero-to-all code generation, where an agent must construct an entire software project directly from natural-language

中文介绍 Repo0 面向 zero-to-all 代码生成，提出设计驱动的全项目生成方法，让 agent 从自然语言需求直接构建完整软件项目，不再依赖预定义仓库架构，解决现有系统在无仓库假设下无法工作的问题。

Looped Language Models Improve Compositional Tool Calling

👍 16

08/17 08:00

Looped language models have shown promising results on reasoning benchmarks, yet their potential for agentic tool use remains largely unexplored. We study this question in compositional tool-calling settings, where models must coordinate multiple API calls, maintain intermediate state, and preserve

中文介绍 Looped Language Models 研究循环语言模型在组合式工具调用中的表现，证明其能更好地协调多次 API 调用、维护中间状态并保持调用一致性，将循环推理能力从纯 reasoning 扩展到 agentic tool use。

FM-Bench: A Benchmark for Long-Horizon Management with Competing Agents

👍 16

08/19 08:00

Language model agents now execute bounded tasks reliably. Whether they can sustain effective decision-making over long horizons, where actions have cumulative consequences and the environment responds to their choices, remains largely unmeasured. FM-Bench (Football Management Benchmark) measures thi

中文介绍 FM-Bench 提出足球管理基准，用于度量语言模型智能体在长时程、有竞争 agent 的动态环境中的决策能力，强调行动具有累积后果且环境会响应选择，弥补现有评测仅覆盖有界任务的不足。

Decision-Metric Alignment in Latent World Models: Diagnostics and Action-Conditioned Objectives for MPC Planning

👍 15

08/19 08:00

JEPA-style latent world models can use Euclidean distance to a goal latent as the cost for model-predictive control (MPC). Strong decoding of task variables, however, does not guarantee that this particular cost ranks candidate action sequences by real task progress. We call the latter property deci

中文介绍 针对 JEPA 类潜空间世界模型中用欧氏距离作为 MPC 代价可能无法按真实任务进展排序动作序列的问题，提出诊断工具与动作条件化目标函数，使决策度量与任务目标对齐，提升规划质量。

FlashPrefill V2: Block-Sparse Prefill Attention for Long-Context LLM Serving

👍 14

08/20 16:02

Long-context modeling is a pivotal capability for Large Language Models, yet the quadratic complexity of attention remains a critical bottleneck, particularly during the compute-intensive prefilling phase. Our previous work, FlashPrefill, mitigates this cost through instantaneous pattern discovery a

中文介绍 FlashPrefill V2 提出块稀疏 prefill 注意力机制，面向长上下文 LLM 服务，通过块稀疏模式降低 prefill 阶段二次复杂度，在保留前代瞬时模式发现优势的同时进一步加速长上下文预填充。

Scaling Creative Writing Beyond Story-Centric Data with Attribute-Guided Genre Expansion

👍 14

08/14 08:00

High-quality creative writing data for large language models (LLMs) remains dominated by story-centric data, limiting models' ability to follow the structural and functional conventions of diverse creative formats. We propose an attribute-guided genre expansion framework for scaling creative writing

中文介绍 提出属性引导的文体扩展框架，突破故事类数据主导的创意写作训练，按属性生成覆盖多种创意格式的高质量数据，使 LLM 学会遵循不同文体的结构与功能惯例，提升多样化创意写作能力。

Thinking in a Low-Resource Language: What SFT Builds, What RL Fixes, What Accuracy Cannot See

👍 13

08/18 21:09

Take three frontier mixture-of-experts models (Alibaba, OpenAI, NVIDIA; 3.6-4.0B active parameters each) and fine-tune them to reason in a low-resource language. On accuracy benchmarks almost nothing happens, and the benchmark itself is noise at this scale: changing only the random seed moves the sc

中文介绍 通过在三款前沿 MoE 模型上用低资源语言微调推理，发现准确率基准几乎不变且受随机种子影响，纯准确率无法捕捉真实变化；对比 SFT 与 RL 各自的作用，揭示 RL 能修复 SFT 构建之外的能力。

VA-Judger: Reward Modeling from Human Preference Feedback for Joint Video-Audio Generation

👍 11

08/19 08:00

Using reinforcement learning to post-train joint video-audio generation models requires a reward signal. Existing methods construct this reward by combining metrics for individual quality dimensions, including audio quality, visual fidelity, and synchronization. However, these metrics evaluate perce

中文介绍 VA-Judger 提出从人类偏好反馈学习奖励模型，用于视频-音频联合生成模型的 RL 后训练；不再简单组合音频质量、视觉保真度与同步性等指标，而是直接建模人类对联合输出的整体感知质量，提供更有效的奖励信号。

Update on our post-training effort

@gabepereyra · 11.4K 粉丝 · 616.8K 阅 · 507 赞 · 99 转

08/20 23:00

Authors: @nikogrupen @ItsJulioPereyra @calvincongelado @vtrengarajan @gabepereyra Over the past six months, Harvey’s research agenda has focused on two goals: Building frontier legal intelligence

中文介绍 Harvey 研究团队发文总结过去六个月研究重点：构建前沿法律智能（frontier legal intelligence）。目标是推进 AI 在法律领域的专业能力，但第二个目标未在预览中提及，信息有限。

Turn Grok Bot Into Your CTO

@RayFernando1337 · 37.8K 粉丝 · 169.9K 阅 · 520 赞 · 85 转

08/20 05:54

I've turned Grok Bot into my CTO and I'm no longer the bottleneck for new ideas, bug fixes, or anything involving the project. Prompt to CTO When working on a new project I create a bot and give it

中文介绍 博主把 Grok Bot 变成自己的 CTO：新建项目时创建 bot 并给它 CTO 角色 prompt，负责新想法、bug 修复等，使自己不再是项目瓶颈。重点分享 prompt 设置方法。

How To Set Up Grokbot The Right Way So It Runs Your Business While You Sleep In 2026

@heynavtoor · 149.3K 粉丝 · 161.9K 阅 · 550 赞 · 93 转

08/20 20:49

Elon Musk just told his followers to try Grokbot after a power user showed it replaced his $10,000 Mac Mini local AI setup (Elon on X, Aug 16). Gavin Baker says his AI usage went up about 100x after

中文介绍 这是 Grokbot 的正确设置教程，让你睡觉时业务也能运行。背景：Elon Musk 曾推荐，有用户用它替代了 1 万美元的 Mac Mini 本地 AI 配置，Gavin Baker 称 AI 使用量提升约 100 倍。

Building software factories (with no slop)

@dzhng · 13.3K 粉丝 · 137.6K 阅 · 502 赞 · 48 转

08/20 09:39

The amount of code being written today far outpaces our ability to review it. That part isn't controversial. What's more interesting is that we're visibly losing — the amount of AI slop in modern

中文介绍 讨论当前代码生产速度远超人工审查能力，AI 生成的低质量代码（AI slop）正在泛滥，我们明显处于劣势。提出建设「软件工厂」并避免 slop 的思路。

Agentic coding is now multiplayer: Introducing Slack Code

@SlackHQ · 434.3K 粉丝 · 86.9K 阅 · 512 赞 · 44 转

08/20 20:28

AI coding is now a team sport. Code channels bring software development out of private tabs, so teams and agents can write, review, and ship code together, in the open. Slack is where agents become

中文介绍 Slack 发布 Slack Code，主打「agentic coding 是团队运动」：代码频道让开发走出私人标签页，团队与 AI agent 可一起编写、审查、交付代码。产品发布。

AI Engineering Skills Map: Building and Deploying AI Applications

@AndrewYNg · 1.8M 粉丝 · 62.4K 阅 · 1.2K 赞 · 189 转

08/22 00:37

I previously wrote about our AI Engineering Skills Map, with the highest level skills being (i) Building and deploying AI applications, (ii) Software engineering fundamentals, (iii) Using coding

中文介绍 Andrew Ng 分享 AI 工程技能图谱：最高层技能是构建和部署 AI 应用、软件工程基础、使用编码工具等。这是对其先前内容的新补充，明确 AI 工程师的能力层级。

RL Policy Churn

@ID_AA_Carmack · 3.7M 粉丝 · 58.8K 阅 · 577 赞 · 40 转

08/21 03:01

I have a good theory about this now. The Phenomenon of Policy Churn https://arxiv.org/pdf/2206.00730 The classic exploration method in RL is “epsilon greedy”, where the agent normally takes the “greedy” action

中文介绍 Carmack 分享强化学习中「策略翻腾」（Policy Churn）现象的理论解释，引用相关论文，并讨论 epsilon-greedy 传统探索方法及其局限。技术分析帖。

Grok Bot vs. Hermes: They Look Similar. Underneath, They’re Making Very Different Bets.

@themahis · 4.7K 粉丝 · 43.0K 阅 · 515 赞 · 37 转

08/20 23:59

One is trying to make AI teammates feel like a managed product. The other is turning the agent itself into an open, composable runtime. Research note — August 20, 2026: For this comparison, I

中文介绍 对比 Grok Bot 与 Hermes：一个将 AI 队友做成受管产品，另一个把 agent 本身做成开放、可组合的运行时。两者表面相似，底层方向截然不同。研究笔记。

5 design patterns for long-horizon agent harness

@GoogleCloudTech · 1.3M 粉丝 · 42.9K 阅 · 506 赞 · 87 转

08/20 09:23

Any agent can look brilliant on a one-shot task. Give it a week of real work and it starts to fall apart. "Long horizon" means the agent keeps going across days and dozens of sessions instead of

中文介绍 Google Cloud 分享长时程 agent 的 5 个设计模式：短期内亮眼的 agent 在跨天、多会话的真实工作中会崩溃，因此需要针对长期运行专门设计的架构模式。

System Design for Agent Systems (Part 1)

@kmeanskaran · 13.9K 粉丝 · 32.9K 阅 · 518 赞 · 66 转

08/20 21:52

Put a frontier model into a badly designed agent system and you get a more articulate failure. Nearly all of the engineering sits outside the model, in the scaffolding that has come to be called the

中文介绍 关于 Agent 系统设计的第一部分：把前沿模型放进设计糟糕的 agent 系统中，只会得到更「清晰」的失败。几乎所有工程都位于模型外部，即 scaffolding（脚手架）层。

[AINews] 10% worse, 100x cheaper, 10000x faster: Why Simulation is taking over

Latent Space · 08/22 15:36

Did you think RSI stopped at model training?

中文介绍 本期AI新闻探讨模拟技术接管趋势：性能仅下降10%，但成本降低100倍、速度提升10000倍。文章指出，模拟的应用并未止步于模型训练阶段。

The Evolution of the Agent Harness

Latent Space · 08/22 15:30

Models keep absorbing the harness into their weights — soon, it will be a harness for human attention rather than for the model.

中文介绍 文章探讨智能体控制机制（harness）的演变：模型不断将外部控制内化进自身权重，未来这种机制或将用于引导人类注意力，而非约束模型。

Simulation: the new Scaling Law — Joon Sung Park, Simile AI

Latent Space · 08/22 07:37

Simile’s CEO about his journey from the viral Generative Agents to creating 8 Billion Digital Twins of every living human... and why it’s gone from fun exploration to very serious business.

中文介绍 Simile AI首席执行官Joon Sung Park讲述从爆款项目Generative Agents到为每个在世人类创建80亿数字孪生的历程，并解释为何这一探索已从趣味尝试转变为严肃商业。

From Atari to EVE Online: Building on 15 Years of AI Research in Games

DeepMind Blog · 08/21 19:59

Google DeepMind partners with game studios to prototype breakthrough AI gameplay.

中文介绍 Google DeepMind宣布与游戏工作室合作，原型化突破性AI玩法，纪念其在游戏领域AI研究15周年，涵盖从Atari到EVE Online的演进。

[AINews] Poolside gets $12B reverse-execuhire to NVIDIA; founders stay for $1B, employees go for $6B, Infraco scaling to 7GW neocloud

Latent Space · 08/21 13:45

Yes, we’re confused too.

中文介绍 Poolside通过一笔120亿美元交易并入NVIDIA：创始人留任获10亿美元，员工获60亿美元，基础设施公司Infraco将扩展至7GW规模的新云。

not much happened today

Smol AI News · 08/21 13:44

**Ox Alpha** emerged as a mystery model with strong coding and agentic performance, likely a **Zhipu/GLM-family** model such as **GLM-5.3 Vision**. Analysts suggest its gains come from post-training and infrastructure improvements rather than sheer size, based on the **743B base** of **GLM-5.2** wit

中文介绍 神秘模型Ox Alpha亮相，编码与智能体任务表现强劲，可能属于智谱GLM系列（如GLM-5.3 Vision）。分析认为其性能提升来自后训练与基础设施改进，而非单纯规模扩大，基础模型为743B。

Measuring benchmark optimization in speech recognition

Hugging Face Blog · 08/21 08:00

中文介绍 Hugging Face博客文章聚焦语音识别中的基准优化测量，旨在探讨如何准确量化该领域的模型改进。

ChatGPT Apple Messages 💬, Anthropic’s meeting recorder 💼, Mistral Agentic Search 🔍

TLDR AI · 08/21 08:00

中文介绍 TLDR AI本期简报：ChatGPT接入苹果Messages，Anthropic推出会议记录器，Mistral发布智能体搜索功能。

The /wayfinder Skill: Navigating the “Fog of War” of Planning

Latent Space · 08/21 04:59

Matt Pocock tells us about his /wayfinder skill, for greenfield projects or for when the way forward is unclear.

中文介绍 Matt Pocock在文章中介绍他开发的/wayfinder技能，该技能用于全新项目或前路不明之时，帮助用户在规划的“战争迷雾”中完成导航。

Up to 3.2x Faster Inference with LFM2.5-DSpark

Hugging Face Blog · 08/21 00:52

中文介绍 Hugging Face博客介绍Liquid AI推出的LFM2.5-DSpark模型，宣称其推理速度相比现有方案最高可提升3.2倍。

Debates over AI consciousness are a trap

MIT Tech Review AI · 08/20 23:42

“Runaway” AI, “rogue” agents, and “autonomous” actors—the current rhetoric would have you believe that AI agents are not only awake and aware, but angry at their creators. Prominent tech leaders such as Demis Hassabis, Dario Amodei, and Sam Altman push for regulation of these seemingly “superhuman”

中文介绍 MIT科技评论文章认为，围绕AI意识的争论是个陷阱。当前“失控AI”“流氓智能体”等修辞暗示AI智能体已觉醒并对创造者愤怒，Hassabis、Amodei、Altman等科技领袖正推动监管。

Unlocking hidden revenue streams with market models

MIT Tech Review AI · 08/20 17:47

Each day, an airline transports tens of thousands of passengers on hundreds of flights. Often these are not straightforward point-to-point routes, with passengers requiring multiple connections. The airline can consider potentially hundreds of variables to price each of these journeys: demand, seaso

中文介绍 文章探讨利用市场模型解锁隐藏收入流。以航空公司为例，其每天运送数万名乘客、执行数百个航班，需考虑数百个变量为多段中转旅程定价。

Introducing AI Futures

OpenAI News · 08/20 15:00

Introducing AI Futures, a new OpenAI blog exploring how transformative AI could reshape power, governance, the economy, and individual freedom.

中文介绍 OpenAI推出名为“AI Futures”的新博客，旨在探讨变革性AI如何重塑权力、治理、经济与个人自由。

not much happened today

Smol AI News · 08/20 13:44

**OpenAI** and **Anthropic** expanded their agent platforms with new desktop features, collaborative editing, and composable APIs like Skills and Files API. **OpenAI** rolled out memory and workflow features in the EEA, UK, and Switzerland. **AT&T** revealed that 40% of employee AI usage routes to o

中文介绍 OpenAI与Anthropic扩展智能体平台，新增桌面功能、协作编辑及Skills、Files API等可组合API；OpenAI在EEA、英国和瑞士推出记忆与工作流功能。

[AINews] Death of Params: Z.ai CEO Jie Tang on GLM 5.3 and the new Post-training Scaling Law

Latent Space · 08/20 13:17

Every lab CEO is on X now

中文介绍 本期AI新闻聚焦“参数的终结”：Z.ai首席执行官唐杰（Jie Tang）就GLM 5.3模型与新的后训练扩展定律发表看法。