2026-08-14

Mechanist: AI as a Scientific Instrument for Discovering the Mechanisms of Intelligence

👍 73

08/12 08:00

AI models have achieved remarkable success across diverse domains, yet the mechanisms underlying their capabilities and the risks they may pose remain poorly understood. As AI development becomes faster and increasingly automated, mechanistic exploration remains largely manual, widening the gap betw

Ready Cohorts: Bounding GPU Opportunity and Avoiding Host Round Trips in LLM-Agent Control

👍 1

08/12 08:00

LLM-agent services repeatedly execute small deterministic transitions between model and tool calls: route an outcome, update state, and emit the next effect. We ask when this control path exposes enough concurrent work for GPU execution, and what changes when a GPU-computed route decision remains on

Hand Visibility Detector: Per-Keypoint Visibility Estimation for Hands

👍 1

08/12 08:00

Hand Pose Estimation (HPE) is a fundamental technology for various applications such as AR/VR and robotics. In these applications, the visibility of each hand joint in the image is crucial for assessing the reliability of estimation results under occlusion. However, most existing HPE methods output

Persistent Recursive Worlds Enable Autonomous Software Evolution

👍 4

08/12 08:00

Complex software systems develop over timescales that exceed the lifespan of any individual coding agent. Most agentic software systems preserve continuity through persistent sessions, memories, managers or shared context. We introduce EvoX Genesis (hereafter, Genesis), which instead makes the softw

AI4AI at Test-Time: Strong-to-Weak Capability Transfer via Harnesses

👍 97

08/12 08:00

Recent work on distillation transfers the capabilities of large models to smaller ones often by updating the latter's parameters, through teacher forcing, on-policy distillation, and related training-time methods. In this paper, we ask whether such transfer can instead occur at test time. We study s

From Synthesis to Removal: Physics-Grounded Reflection Simulation and Diffusion-Based Video Dereflection

👍 6

08/12 08:00

Videos captured through glass often contain reflections that degrade visual quality and interfere with downstream vision tasks. Although single-image reflection removal has been extensively studied, video reflection removal remains largely underexplored due to the lack of paired video data, temporal

MBA: Multimodal Benchmark and Agents for Real-World Business Ideation

👍 5

08/12 08:00

Agentic systems powered by large language models (LLMs) have opened new opportunities for business ideation. Yet existing approaches remain confined to a text-only paradigm, despite the inherently multimodal nature of real-world contexts. We thus introduce MBA-Bench, the first multimodal benchmark f

Spark-to-Paper: End-to-End Research Paper Generation as a Composable Skill

👍 176

08/12 08:00

Turning a research idea into a complete paper requires more than text generation: the system must retrieve literature, design and execute experiments, revise claims according to evidence, produce publication-ready figures, and maintain consistency across a long generation process. We present Spark-t

ToolHazard: Scaling Adversarial Environments for Security Evaluation and Alignment of LLM-based Agents

👍 7

08/12 08:00

Large language model (LLM) agents integrated with external tools are vulnerable to indirect prompt injections embedded in environmental states. However, existing studies largely rely on manually implemented or reused environments, stochastic LLM-based tool simulation, and predefined injection locati

ReRound: Reconstructive Rounding to Resolve Midpoint Ambiguity in Calibration-Free LLM Quantization

👍 0

08/11 08:00

ReRound (Reconstructive Rounding) is a post-training quantization method that addresses the midpoint ambiguity inherent in standard round-to-nearest (RTN) schemes when quantizing weights near the centers of quantization intervals. Starting from a pretrained LLM, ReRound trains a conditional diffus

Self-Evolving Embodied Agents via Skill-Harness Evolution

👍 9

08/11 08:00

Embodied agents are increasingly built as systems around foundation models, where performance depends not only on model weights but also on the skills, context, action interfaces, and execution harness surrounding the model. While supervised fine-tuning and reinforcement learning can adapt agents to

Agent Safety Should Be a Runtime Contract

👍 3

08/11 08:00

The dominant paradigm treats AI safety as a property to be instilled during model training via RLHF, DPO, or Constitutional AI. We argue this is structurally insufficient for autonomous agents that execute code, mutate files, send messages, and modify databases. Agent safety should be a runtime cont

InSight-doc: Agentic Visual Perception for Long-Document Understanding

👍 11

08/11 08:00

Long-document understanding often requires reasoning over many visually rich pages, making inference costly and prone to context rot. In this work, we propose InSight-doc, an agentic visual perception framework that treats visual resolution as an adaptive reasoning-time resource. InSight-doc starts

DistilVDR: A Compact End-to-End Visual Document Retriever via Dual-Student Distillation

👍 11

08/11 08:00

Visual document retrieval (VDR) is dominated by multi-billion-parameter models that are slow to index at full corpus scale and expensive to serve. Prior compression routes either train a smaller multi-vector encoder from scratch or distil only the query side; neither yields a compact single-vector r

Reference-Free Post-Training of Open Large Language Models for Multilingual Machine Translation

👍 12

08/11 08:00

We study reference-free post-training for multilingual machine translation with open large language models. Starting from the supervised-finetuned MiLMMT-46-v0.1 models, we apply Group Relative Policy Optimization (GRPO) with a reward that averages two reference-free quality estimation models and is

Parameter Exploration for RLVR via Variational Learning

👍 4

08/11 00:28

Exploration has been a focus of reinforcement learning research for a long time. Recently, there has been growing evidence that it is also an important ingredient in LLM reinforcement learning recipes that can significantly impact downstream performance. Many existing methods control exploration in

Learning How the World Evolves: Extrapolative Video World Models via Latent Dynamics Reasoning

👍 6

08/10 08:00

The world evolves following its dynamics, i.e., its laws of motion. However, leading video diffusion models largely fit the pixels without modeling how the pixels transit over time. Thus, they render visually plausible frames but may not accurately obey the laws. To capture the dynamics purely from

Power law graph attention: exact generalization of scaled dot-product attention, empirical collapse at inference

👍 7

08/10 08:00

The Large Language Model from Power Law Decoder Representations (PLDR-LLM) and its attention, Power Law Graph Attention (PLGA), replace the fixed bilinear form of scaled dot-product attention (SDPA) with a learned, input-generated bilinear operator G_{LM}, built from a positive tensor A_{LM} by elem

Decoding-Level Taboo: A Diagnostic Stress Test for LLM Robustness

👍 13

08/10 08:00

Large language model evaluations typically focus on performance under nominal conditions, creating an illusion of capability where models comfortably walk a narrow, highly optimized generation corridor. In real-world deployments, however, complex system prompts, safety guardrails, and structural con

360CityArena: A Realistic Virtual Urban Navigation Benchmark for Embodied Agents

👍 10

08/09 08:00

We present 360CityArena, a benchmark for evaluating the urban exploration capabilities of embodied agents within a photorealistic environment constructed from 360-degree videos. Existing outdoor benchmarks either lack sufficient photorealism or complexity, resulting in a considerable gap from real-w

Not Worth Another Token: Marginal Value Estimation for Efficient Deep Research Agents

👍 12

08/09 08:00

Long-horizon research agents solve open-ended tasks through iterative retrieval, aggregation, and synthesis, but context grows rapidly while the marginal value of additional evidence often declines. This leads to unnecessary token cost, higher latency, and noisier inputs for final report generation.

Can LLM Agents Stick to the Script? A Benchmark for Long-Horizon Consistency in Interactive Narratives

👍 25

08/08 08:00

The rapid advancement of Large Language Models (LLMs) is revolutionizing AI for Games by enabling open-ended and fluid interactive storytelling. However, existing research has largely overlooked the critical challenge of maintaining long-horizon logical consistency and narrative integrity against un

NeuPAT: Neuron-aware Plasticity Allocation Tuning for Language-Preserving MLLMs

👍 2

08/08 08:00

Multimodal expansion of large language models (LLMs) enables new perceptual capabilities but often compromises the language intelligence acquired during pretraining. In this work, we investigate this phenomenon from the perspective of internal adaptation dynamics and discover that neurons in pretrai

AtlasVLA: Persistent World-Ego State Modeling for Vision-Language-Action Models

👍 2

08/07 08:00

While Vision-Language-Action (VLA) models have advanced embodied AI, their fundamentally reactive paradigm severely limits performance in partially observable and long-horizon tasks. When restricted to a single wrist-mounted camera, they inevitably suffer from perception forgetting as objects exit t

CoinRAG: Contextualized Information Nugget KV Cache Reuse for Long-Context RAG

👍 6

08/07 08:00

Recent optimization studies on Retrieval-Augmented Generation (RAG) have exploited chunk-level KV cache reuse to avoid processing long retrieved contexts for higher efficiency, while significant information redundancy and noise still remain in the coarse-grained chunks. This paper optimizes the Pare

SkillZip: Contract-Preserving Graph Compression for Scalable Agent Skill Libraries

👍 72

08/06 08:00

Large Language Models (LLMs) increasingly act as agents whose procedural knowledge is stored in reusable skill packages and loaded at inference time. As skill libraries grow, a central challenge is to expose the smallest sufficient executable context under a limited context budget. Existing systems

The Illusion of Visual Tool-Use: A Causal Audit of Thinking with Images

👍 6

08/06 08:00

The "thinking-with-images" paradigm equips multimodal LLMs with active visual operations such as crop-and-zoom. However, models using these operations often achieve only marginal or negative gains over direct inference at substantially higher token cost. They may also repeatedly crop irrelevant regi

OpenART: Scaling Agent Red Teaming via Open-Ended Environment Evolution

👍 182

08/01 08:00

AI agents operate in persistent environments where early state changes can influence decisions far into the future. Unlike conventional language-model interactions, agent behavior is mediated through a shared state that is repeatedly modified and reused across long-horizon workflows. Current safety

AutoWorldModel-Bench: A State-Centric Benchmark for Automated World-Model Research

👍 10

07/20 08:00

World modeling is an unsettled field: architectures, training objectives, and state representations interact in complex ways, and no single recipe dominates across environments. This makes it an ideal testbed for AI coding agents acting as autonomous researchers--a setting in which the improvement d

Poor Man's Agentic Modeling: Simulating Large LLM-Agent Societies on a Laptop

👍 2

07/19 08:00

Simulating societies of many large language model (LLM) agents is expensive, yet the questions asked of such simulations are usually macroscopic: phase behaviour, stylised facts, and scaling with the number of agents N, not the cognition of any single agent. We turn a statistical-physics observation

To FDE, or not to FDE?

@thejessezhang · 85.7K 粉丝 · 270.4K 阅 · 510 赞 · 30 转

08/11 23:24

Two-thirds of our deployment work is now done autonomously by our own product (via Duet). This post is about why that is, our philosophy on building a product + deployment motion, and how that's

中文介绍 探讨自主部署哲学，Duet 产品使三分之二部署工作自动化。

Agentic Code Quality

@addyosmani · 408.3K 粉丝 · 175.5K 阅 · 509 赞 · 65 转

08/12 14:36

For much of human history, we've evaluated code quality via code review: someone reads what you wrote and makes sure it's clean, thoughtful, fast, understandable, and tests well. For agents, that

中文介绍 探讨通过代码审查评估代码质量的传统方法，及其在智能体中的应用。

NVIDIA Nemotron 3.5 Lightning and NeMo Switchyard Deliver Faster, Smarter, More Efficient Agentic AI

@nvidia · 2.6M 粉丝 · 59.5K 阅 · 548 赞 · 66 转

08/11 21:41

The new lightweight open model and routing library delivers greater control over AI, data and workflows across edge devices, PCs, workstations, data centers and the cloud. Source: NVIDIA Blog By Kari

中文介绍 NVIDIA 推出 Nemotron 3.5 和 NeMo Switchyard，提升边缘设备到云端的 AI 控制效率。

Agent Plugins are the future of Agent Skills

@GoogleCloudTech · 1.3M 粉丝 · 56.7K 阅 · 505 赞 · 76 转

08/13 10:49

Agent Plugins is an open, vendor-neutral standard for packaging Agent Skills and the MCP servers they depend on into one portable folder that any compatible client can load. Google is joining the

中文介绍 Agent Plugins 成为智能体技能打包与加载的标准，Google Cloud 加入支持。

/show-me: compact visual representations for coding agents

@dexhorthy · 30.0K 粉丝 · 42.9K 阅 · 635 赞 · 31 转

08/12 23:59

tl;dr make your agent converse visually instead of in walls of prose. Lighter and faster than HTML, good enough for most dev-work shaped problems. Coding agents are pretty much unreadable The

中文介绍 介绍 /show-me 工具，为编码智能体提供视觉对话方式，提升开发效率。

Intro to Grok Bot

@mattyp · 44.5K 粉丝 · 33.5K 阅 · 651 赞 · 48 转

08/12 02:59

I’m always chasing better tools. Notes apps, workflows, optimizations, and now, personal agents. It’s an easy trap to fall into - always building the “custom” thing and sacrificing the work as a

中文介绍 介绍 Grok Bot，分享构建个性化工具的陷阱与优化心得。

Grok 4.6 – A field guide

@ericzakariasson · 80.4K 粉丝 · 32.2K 阅 · 719 赞 · 52 转

08/12 23:46

Grok 4.6 is out! I've used it for a few weeks as my daily driver across the normal mix of coding and knowledge work, and built a few projects with it specifically to push on where it holds up. It's

中文介绍 Grok 4.6 发布，分享作为日常驱动工具和项目开发的体验。

Introducing Gemini 3.7 Flash

@GoogleAIStudio · 191.7K 粉丝 · 30.7K 阅 · 566 赞 · 60 转

08/14 01:25

Today, we’re building on the progress of our widely used Flash series by introducing Gemini 3.7 Flash, our most intelligent workhorse model yet for coding and agents. This release comes just three

中文介绍 Gemini 3.7 Flash 推出，成为编码和智能体工作最智能的模型。

Record, train, and deploy from one place with Strands Agents, LeRobot, and Hugging Face Storage Buckets

Hugging Face Blog · 08/14 01:16

中文介绍 Hugging Face Blog介绍Strands Agents、LeRobot和Hugging Face Storage Buckets，实现数据记录、训练和部署一体化。

Introducing Gemini 3.7 Flash

DeepMind Blog · 08/14 01:04

中文介绍 DeepMind Blog发布Gemini 3.7 Flash，引入新的模型和功能。

Flock is tightening its rules in response to a growing surveillance backlash

MIT Tech Review AI · 08/13 21:41

The police-tech giant Flock is announcing today that it will change officers’ access to its nationwide network of license plate readers, in an apparent effort to quell a growing backlash and win back contracts lost amid concerns about mass surveillance and police abuse. Several changes aim directly

中文介绍 MIT Tech Review AI报道，Flock针对日益增长的监控反感和警察滥用问题，调整全国车牌识别网络的警员访问权限。

The builder’s guide to GPT‑5.6

OpenAI News · 08/13 19:00

Learn how startups use GPT-5.6 to build faster, more cost-efficient AI agents with smarter model selection and new Responses API capabilities.

中文介绍 OpenAI发布GPT-5.6构建指南，帮助初创公司利用GPT-5.6构建更高效的人工智能代理。

Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speed

OpenAI News · 08/13 18:00

Preview Ultrafast, a new OpenAI API service tier that runs GPT-5.6 Sol up to 14× faster. Powered by Cerebras, it delivers up to 750 output tokens per second.

中文介绍 OpenAI推出Ultrafast模式，GPT-5.6 Sol速度提升至14倍，由Cerebras提供支持，每秒输出高达750个token。

OpenAI appoints Dali Rajic as Chief Revenue Officer

OpenAI News · 08/13 17:00

OpenAI appoints Dali Rajic as Chief Revenue Officer to lead its global revenue organization and help businesses realize the full value of AI.

中文介绍 OpenAI任命Dali Rajic为首席收入官，领导全球收入组织，帮助企业实现AI的最大价值。

How kids feel about AI, in their own words

MIT Tech Review AI · 08/13 17:00

When we set out to talk to kids about artificial intelligence, we thought we knew what we’d hear. We expected some to tell us they were using it to cheat a little, the way Millennials and Gen Xers opened up CliffsNotes or programmed formulas into their TI-82s, and others to share inspiring ways they

中文介绍 MIT Tech Review AI探讨孩子们对人工智能的看法，了解他们对AI的应用和感受。

[AINews] SpaceXAI Grok 4.6 and Grok @Bot

Latent Space · 08/13 09:53

AI teammate category just had its most significant new entrant yet

中文介绍 Latent Space报道，AI团队成员类别迎来重大新成员。

What We Learned by Reproducing 2,200 papers from ICML

Hugging Face Blog · 08/13 08:00

中文介绍 Hugging Face Blog介绍通过复现ICML 2200篇论文所获得的经验。

Claude Chrome Cowork 🌐, Grok 4.6 🚀, DeepSeek v4-Pro-0813 🧠

TLDR AI · 08/13 08:00

中文介绍 TLDR AI报道Claude Chrome Cowork、Grok 4.6和DeepSeek v4-Pro-0813等AI产品。

Scaling AI agents with trustworthy data

MIT Tech Review AI · 08/13 00:51

Business and technology leaders need no convincing that the time of agentic AI is here. Organizations are rapidly adopting agents, and few executives doubt the technology’s potential to transform work. But many organizations find that realizing the desired return on investment (ROI) from AI hinges o

中文介绍 MIT Tech Review AI探讨如何通过可信数据扩大AI代理规模，实现投资回报。

Introducing OlmoEarth embeddings: Custom embedding exports from OlmoEarth Studio for downstream analysis

Hugging Face Blog · 08/13 00:14

中文介绍 Hugging Face Blog介绍OlmoEarth Studio的定制嵌入导出功能，用于下游分析。

Putting sign language AI into users’ hands

DeepMind Blog · 08/12 22:01

Introducing sign-language-to-text (SL2T), our breakthrough model powering new sign language features for Deaf and hard of hearing users.

中文介绍 DeepMind Blog推出手语到文本（SL2T）模型，为听障和听力受损用户提供新的手语功能。

LFM2.5-VL-3B for Better and Faster Vision Capabilities for the Edge

Hugging Face Blog · 08/12 22:00

中文介绍 Hugging Face Blog介绍LFM2.5-VL-3B，用于边缘设备的更高效和更快的视觉能力。

[AINews] How to steal a Reasoning Trace

Latent Space · 08/12 15:11

Speculative Decoding by any other name would distil as sweet

中文介绍 Latent Space报道如何窃取推理跟踪，也称为推测解码。