2026-10-15

Judges Follow the Reference: The Derivation Gap in Verifying Tool-Call Arguments

👍 0

10/13 08:00

Tool-call judges evaluate argument values proposed by LLM agents. We find that they accept a correct value less often when their input holds no reference for it, that is, no text that says what the value should be. This happens when the value is a default value from the function schema that the user

Beyond the Verdict: Agentic Code Judging with Verifiable Feedback

👍 4

10/13 08:00

Test-time scaling for large language models (LLMs) can involve generating multiple candidate solutions for selection or revising an earlier attempt. To guide selection and revision, we need a verifier that can estimate candidate correctness and provide evidence of failure. In competitive programming

Enabling Token-wise Importance Sampling in Off-Policy RLVR with Proximal KL Regularization

👍 1

10/13 08:00

Off-policy data in reinforcement learning with verifiable rewards (RLVR) can arise from asynchronous generation, rollout-training mismatch, and response reuse. Token-wise importance weighting avoids multiplying likelihood ratios across a response but generally biases policy gradients. We prove that,

Reading Right, Acting Wrong: Diagnosing Decision Failures of System One Models in Agent Loops

👍 5

10/13 08:00

System One models choose an agent's next action by scoring a supplied set of options in a single pass. This makes them fast and cheap, but when they choose wrongly there is nothing to inspect: did the model misread what the agent has seen, or read it correctly and still act wrongly? We answer behavi

Can an Agent Learn to Predict Its Own Failure? Outcome-Grounded Confidence Calibration in LLM Agents

👍 1

10/13 08:00

Controlling an acting LLM agent means deciding, at each step of a long-horizon task, whether to trust its next action, abstain, or hand off to a human. That decision needs a per-step confidence signal predicting whether the current rollout will eventually succeed. The agent's own self-reported verba

Let the Mind Conceive, Let the Hand Achieve: Following Multimodal Instructions for Long-Horizon Robotic Manipulation

👍 1

10/13 08:00

A robot following a multi-step instruction needs to act on the right objects in the right order, even as the scene changes. It must also maintain awareness of task progress and determine when to advance. To address these challenges, we introduce Orchestra, an agent harness that separates task planni

LifeSearch: Benchmarking Deep Research Agents for Real-World Life and Productivity

👍 15

10/13 08:00

Deep research (DR) agents address open-ended information needs through multi-step reasoning, long-horizon web search, and evidence synthesis. However, existing benchmarks primarily focus on academic research or synthetic tasks, leaving DR agents' capabilities in real-world life and professional prod

SHELDON: Reasoning through Dual Search over Hypothesis and Observation Spaces

👍 1

10/13 08:00

Test-time scaling has enabled LLMs to solve increasingly complex reasoning, discovery, and optimization tasks. However, existing approaches often gather observations while seeking evidence for current hypotheses, then generate new hypotheses conditioned on those hypotheses and their associated evide

4LC: 4D Latent Cache for Efficient Conditioning in Dynamic World Modeling

👍 1

10/13 08:00

For video generation in dynamic world modeling, observed static and dynamic content should be efficiently reused to provide conditions for repeated generation at different times and viewpoints. We introduce 4LC, a 4D latent cache with composite visual representations and a flexible spatiotemporal ac

SWE-MISBEHAVIOR: Benchmarking and Mitigating Boundary Violations in Software Coding Agents

👍 9

10/13 08:00

We present SWE-MISBEHAVIOR, the first taxonomy and evaluation framework for diagnosing boundary-violating misbehaviors in coding agents. It contains 7.4K human-in-the-loop synthesized samples spanning three layers of agent action space: (i) Code Generation, (ii) File Management, and (iii) System Int

Omni-Modal LLMs Know More Than They Say

👍 2

10/12 08:00

Omni-modal large language models (Omni-LLMs) extend LLMs to process heterogeneous modalities (e.g., vision, audio, and IMU signals) alongside language, enabling a broader range of multimodal tasks. Yet we find that these models often fail to express information from non-textual modalities even when

MixLID: Token-Level Language Identification in Mixed-Language Text

👍 1

10/12 08:00

Processing multilingual text requires both identifying a broad range of languages and distinguishing the language of each token in inputs containing multiple languages. We introduce MixLID, a token-level language identification model supporting 111 language-script combinations. MixLID follows a two-

AUGUstus: Autonomous Driving VLA Model with Unified Generation and Understanding

👍 10

10/12 08:00

Vision-language-action (VLA) models show strong potential for end-to-end autonomous driving by leveraging world knowledge and understanding to excel in long-tail scenarios. However, current VLAs have two limitations: 1) Degeneration into vision-action (VA) models. For low latency, regression/flow-ba

EventLens: Event-Structure Reinforcement Learning for Video Understanding

👍 1

10/12 08:00

Recent multimodal large language models (MLLMs) achieve strong results on video question answering, yet often fail to recover the temporal structure that makes a video coherent. They may merge neighboring events, miss single-event progression, or infer implausible relations across events. We study s

LongR: Unleashing Long-Context Reasoning via Reinforcement Learning with Dense Utility Rewards

👍 13

10/12 08:00

Reinforcement Learning (RL) has emerged as a key driver for LLM reasoning. This capability is equally pivotal in long-context scenarios---such as long-dialogue understanding and structured data analysis---where the challenge extends beyond consuming tokens to performing rigorous deduction. While exi

Self-Distillation with Verbal Regularization

👍 29

10/12 08:00

On-policy self-distillation (OPSD) trains a large language model using supervision from a teacher instantiated from the same model as the student. The teacher is conditioned on privileged information (PI), such as verified solutions. The student, which does not observe PI, learns from the teacher's

Data-Centric Recursive Self-Improvement for Navigation via Policy-Synchronized Language-Action Dual Bootstrapping

👍 25

10/11 08:00

Vision-language navigation requires agents to follow natural-language instructions to reach target locations, making high-quality language supervision essential. Since human annotation is costly, prior work has explored scalable data synthesis methods. However, existing methods typically either use

Who to Trust in Multi-Robot Systems: Joint Robot and Object Hypothesis Trust Inference via Neural-Symbolic Cross-Validation

👍 1

10/10 08:00

Multi-robot systems (MRS) offer significant advantages over single-robot platforms by enabling information sharing and coordination for different tasks in complex environments. However, reliable cooperation critically depends on inferring trust over both information sources and their reported hypoth

Words to World: Benchmarking LLMs in Interactive Physical Environments

👍 0

10/10 08:00

Words are increasingly becoming an interface for action: large language models (LLMs) can now invoke tools, operate devices, and make decisions that affect the physical world. However, current benchmarks largely evaluate LLMs through static text responses, offline tool calls, or simulated tasks, lea

Last Secure Code Benchmark: Is the Code You Vibe Secure?

👍 42

10/10 08:00

AI coding agents now write much of the code that ships, yet whether that code is secure remains unclear. We introduce the Last Secure Code Benchmark, a comprehensive evaluation of secure code generation by coding agents. Built from 150 real vulnerabilities, its 450 tasks span seven widely used langu

What Really Works for Universal Multimodal Embeddings? A Controlled Study of Contrastive Training Strategies

👍 10

10/10 08:00

Contrastive learning has become the standard paradigm for multimodal representation learning, aligning semantically matched pairs while separating mismatched ones in a shared embedding space. Although various contrastive training strategies have been proposed across negative selection, batch constru

EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment

👍 51

10/10 08:00

Determining whether an image is AI-generated does not establish whether the situation it depicts is factually supported. We formulate Image Factuality Assessment: given an image and a generic verification instruction, a system returns a supported or refuted verdict with an evidence-grounded explanat

LightVAE: Towards Compact and Efficient Video Autoencoders

👍 15

10/10 08:00

Latent diffusion models enable high-quality video generation, but video VAE encoding and decoding remain computationally expensive. Existing methods use stage-specific lightweight operators, yet either constrain spatial and temporal computation jointly or remove temporal computation altogether, with

Privacy-Preserving Intent Fulfilment and Assurance for 6G RAN

👍 0

10/09 08:00

Intent-based network management automates 6G service lifecycles, but existing fulfilment and assurance approaches require traffic inspection, per-flow tracking, or vendor-internal telemetry. These requirements conflict with privacy constraints in multi-tenant networks and vendor opacity in O-RAN dep

Video-Arena: A Scalable Agentic Video Benchmark for Spatiotemporal Capability

👍 7

10/09 08:00

Understanding and tracking a dynamic world are fundamental to visual intelligence. Existing video benchmarks provide limited insight into whether models can maintain a coherent understanding of evolving world state. Building such a benchmark from real-world videos is challenging because the underlyi

Learning Transferable World-Action Models from Task-Paired Human Videos

👍 1

10/09 08:00

World-action models inherit broad visual and physical priors from video pre-training. However, standard within-video next-frame prediction allows them to rely on appearance shared between the context and future, rather than learning interactions that transfer across scenes and embodiments. We introd

Latent World Action Model for GUI Agents

👍 12

10/09 08:00

Pixel-space world models have recently emerged as a promising approach for GUI agents, predicting future screens prior to execution and subsequently re-perceiving them for action selection. However, this two-stage paradigm faces two key challenges in GUI environments: (L1) the high computational cos

Pristine Models

👍 11

10/09 01:58

Training on broad, unfiltered data enables high-quality image generation, but can also teach models concepts that should not be generated; restricting the training data ensures safety, but reduces quality because examples containing unsafe concepts also provide useful knowledge that is not itself un

Scaling Verifiable Environments for Long-horizon Work Agents

👍 1

10/08 08:00

Work agents operate over digital artifacts to execute professional knowledge-intensive work, requiring training environments that support long-horizon interaction and trustworthy verification. However, hand-crafted environments incur prohibitive engineering overhead that prevents environment scaling

EMHO: EMbodied Agent Harness Optimization via Experience Traces

👍 10

10/06 08:00

Improving embodied agents often focuses on optimizing the underlying model through training, while the surrounding agent harness that controls planning, context, and tool use is typically engineered. We ask whether this harness can instead improve itself directly from experience traces under sparse

How Grok Bot Changed the Way I Work

@melvindvivas · 6.3K 粉丝 · 209.9K 阅 · 541 赞 · 83 转

10/13 13:20

I've been using AI coding assistants like Claude Code, Codex, and Cursor for quite some time. They've already changed how I build software, but Grok Bot has changed something else: how I work

中文介绍 分享使用 AI 编码助手 Claude Code、Codex 和 Cursor 的体验，特别是 Grok Bot 如何改变工作方式。

Introducing Glass: Learn Things Once.

@eglyman · 393.9K 粉丝 · 88.6K 阅 · 724 赞 · 58 转

10/13 22:00

Last week, David Senra asked me: “How does Ramp move so fast?” Today we’re launching the answer. But before we do, I’d like to tell you why we built it. Giving every employee AI — the best models, the

中文介绍 介绍 Glass 系统，强调通过 AI 为每位员工提供最佳模型，提升工作效率。

35+ UNCENSORED AI SITES FOR 18+ IMAGE AND VIDEO GENERATION

@rexsooo · 8.5K 粉丝 · 83.4K 阅 · 703 赞 · 46 转

10/13 00:52

Adult AI image and video generators, NSFW character chat, anime tools, and AI companions This list covers 35 uncensored or NSFW-capable AI platforms across image generation, video creation, anime, AI

中文介绍 列出 35 个成人 AI 图像和视频生成平台，涵盖动漫工具和 AI 伴侣。

How to make a good RL task

@GottliebEli · 809 粉丝 · 28.9K 阅 · 506 赞 · 37 转

10/14 02:36

I’ve been making lots of tasks lately. Many poor, some decent, and now finally, a few good. Through this work I’ve gained what I believe to be valuable insights into the task generation process that

中文介绍 探讨如何制作好的强化学习任务，分享任务生成过程中的见解。

Gemini 4 Argon available on Google Cloud, built to help solve your business’s hardest problems

@googlecloud · 573.7K 粉丝 · 21.2K 阅 · 502 赞 · 70 转

10/14 00:07

Last week we announced that Gemini Enterprise has evolved to make it even simpler to get work done, and introduced the Gemini agent, your new universal agent for work. But model intelligence underpins

中文介绍 介绍 Gemini 4 Argon 在 Google Cloud 上可用，旨在解决企业最棘手的问题。

Gemini 4 Argon: Built to solve your hardest problems

@GoogleAIStudio · 208.3K 粉丝 · 19.6K 阅 · 635 赞 · 45 转

10/14 01:33

Gemini 4 Argon is now globally available today for developers and enterprises via the API in AI Studio and Gemini platform, for Google AI Ultra subscribers in Google Antigravity, and Google AI Ultra

中文介绍 Gemini 4 Argon 全球发布，为开发者和企业提供解决方案。

How to Build an AI Hacking Agent That Finds Bugs While You Sleep

@zack0x01_ · 5.3K 粉丝 · 19.3K 阅 · 501 赞 · 71 转

10/13 14:52

I built an AI agent that hunts bugs while I sleep. Not a chatbot that gives me advice about hacking, an agent on my own server with hands. It maps a target, tests the bug classes I actually care

中文介绍 构建 AI 漏洞挖掘代理，实现睡眠时自动寻找漏洞。

Crossing the Threshold: How OpenAI and Anthropic Approach Autonomy and Safety in Financial Agents

Latent Space · 10/15 05:40

At AI Engineer New York, frontier labs and financial firms explored how to give agents useful autonomy while keeping consequential decisions under control.

中文介绍 AI工程师纽约前沿实验室和金融公司探讨如何在保持关键决策控制的同时赋予代理有用自主性。

Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio

Latent Space · 10/15 04:35

How Leash Bio tamed DELs, the assay that traumatizes everyone, and used the data to catch cheating models. And why that same data is perfect for AI agents learning to think like chemists.

中文介绍 Leash Bio通过DELs测试数据捕捉作弊模型，并利用这些数据训练像化学家一样思考的AI代理。

HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost

Hugging Face Blog · 10/15 00:12

中文介绍 HiRO-ACEv2以较低成本模拟极端天气，分辨率达到3公里。

Amp puts AI agents to work beyond code with OpenAI models

OpenAI News · 10/14 20:00

See how Amp pushes the frontier of what agents can do with GPT-6 Astra, Ultrafast, GPT-Live-1, and Sign in with ChatGPT.

中文介绍 Amp利用OpenAI模型将AI代理应用于代码之外的领域，如GPT-6 Astra、Ultrafast、GPT-Live-1和ChatGPT。

SentinelOne cuts malware analysis from weeks to hours

OpenAI News · 10/14 20:00

With OpenAI Daybreak, SentinelOne cuts weeks of malware analysis to hours, creating capacity to investigate more threats.

中文介绍 SentinelOne利用OpenAI Daybreak将恶意软件分析时间从数周缩短至数小时。

How Third Moment Research runs 10,000 agents for 36+ hours

OpenAI News · 10/14 15:00

With GPT-6 Astra and GPT-6.1 Sol, Codex, and Daybreak for cybersecurity, this AI lab builds agents that focus for days and improve with use.

中文介绍 Third Moment Research使用GPT-6 Astra和GPT-6.1 Sol等工具运行10,000个代理超过36小时。

[AINews] AI Engineer NYC: Gemini 4 Argon, Vals AI Benchmarks

Latent Space · 10/14 13:40

Gemini 4 Argon went GA today, but we already gave it title story 2 weeks ago and most benchmarks, like CursorBench, simply independently confirm their claims.

中文介绍 Gemini 4 Argon正式发布，此前CursorBench等基准测试已独立证实其性能。

Google DeepMind and Singapore Government collaborate to build fine-tuned AI model for scam detection

DeepMind Blog · 10/14 10:00

Learn how Google DeepMind and GovTech Singapore uses lightweight on-device Gemma models to intercept conversational scams in private chats with zero cloud data egress.

中文介绍 Google DeepMind与新加坡政府合作，开发用于诈骗检测的定制AI模型。

The EU’s next chapter in AI is agentic

OpenAI News · 10/14 08:00

New analysis shows how agentic AI adoption is growing across EU enterprises and small businesses—and where deeper use could unlock more value.

中文介绍 欧盟AI新篇章聚焦于代理型AI的采用，分析显示其在企业和中小企业中增长，更深层次的应用可释放更多价值。

OpenAI dev arrested 🚨, Pi API 🤖, OpenCode 2.0 🧑‍💻

TLDR AI · 10/14 08:00

中文介绍 OpenAI开发者被捕，Pi API和OpenCode 2.0发布。

Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics

Latent Space · 10/14 06:25

Our first robotics pod with Dyna’s Chief Scientist on their big bet on World Action Models & the Next Frontier of Physical AI

中文介绍 Dyna Robotics推出Dyna-2，具有一小时自主性的通用物理代理。

Developing Gemini 4 Argon responsibly and safely

DeepMind Blog · 10/14 03:43

中文介绍 DeepMind负责任和安全地开发Gemini 4 Argon。

Run fast, private agents on-device with LFM2.5-24B-A2B

Hugging Face Blog · 10/14 01:05

中文介绍 在设备上运行快速、私有的LFM2.5-24B-A2B代理。

Gemini 4 Argon for Robotics

DeepMind Blog · 10/14 00:01

中文介绍 DeepMind推出Gemini 4 Argon，用于机器人技术。

Gemini 4 Argon: Built to solve your hardest problems

DeepMind Blog · 10/14 00:01

中文介绍 Gemini 4 Argon：专为解决最复杂问题而设计。