2026-10-22

Training the Whole Agent, Not Just the Model

👍 4

10/20 08:00

Training coding agents typically updates the policy while keeping the harness fixed. During training, trajectories reveal flawed repair strategies and interaction problems that vary across tasks and change as the policy learns. We introduce HarnessAdapt-RL, which trains the policy and adapts its har

World Models with Configurable Perception

👍 1

10/20 08:00

Real-world observations are noisy and often contain excessive task-irrelevant information. Effective world modeling and planning need to abstract observations into world state representations while filtering out irrelevant information. However, encoders in existing world models are typically non-ada

Self-Credit: Credit Assignment for Long-Horizon Agentic RL via Intra-group Rescaling

👍 4

10/20 08:00

LLM agents are commonly trained with RLVR, but the sparse nature of typical RLVR rewards makes credit assignment challenging, especially for long-horizon tasks. We propose Self-Credit, an end-to-end agentic RL method that creates dense supervision signals from sparse outcome rewards using self-gener

HRoPE: Homography Relative Positional Encoding

👍 3

10/20 08:00

Geometric Relative Positional Encoding has become the primary mechanism for injecting spatial inductive biases into multi-view transformers. Existing designs condition attention on camera-level frustums or ray-marched depth, associating cross-view tokens through the sensor geometry. Such sensor-cent

OverSeer: Multi-Agent World Models Beyond What Agents See

👍 4

10/20 08:00

Despite the success of video world models, extending them to multiple agents is difficult because it requires generating several agent-specific video streams that are consistent in time and between agents. To address this gap, we introduce OverSeer, a multi-agent world model that augments egocentric

Evolving Agent Interfaces

👍 17

10/20 08:00

Computer-use agents are limited by the interfaces humans give them. An agent may type a spreadsheet cell by cell when a single line of code would do, or print all two thousand lines of a file through a shell to find one function. Both the interfaces and the knowledge of when to use each are built by

Specter: Fast MoE Inference via Quantization-Enabled Multi-Step Speculative Offloading

👍 19

10/20 08:00

Mixture-of-experts (MoE) models often exceed a single accelerator's memory capacity. Offloading experts to host memory creates a severe PCIe I/O bottleneck because data-dependent routing keeps transfers on the critical path. We present Specter, an inference system that uses quantization to combine s

Harness Self-Evolution for Long-Horizon Terminal Agents

👍 10

10/20 08:00

Long-horizon terminal agents may solve a task in one run yet fail across repeated runs. Stochastic errors can propagate through actions and context, creating a gap between capability and reliable execution. We present RiseHarness, a framework for iteratively improving agent harnesses with model weig

DecisionPivot: A Diagnostic Benchmark for Decision Models in LLM Systems

👍 14

10/20 08:00

Decision models support routing, guardrails, and output verification in LLM systems. However, item accuracy alone cannot establish whether models make correct decisions across controlled variations or assign reliable confidence to their predictions. We introduce DecisionPivot, a diagnostic benchmark

VideoMaMa++: Temporally Consistent Video Matting via Preserve-and-Refine Token

👍 1

10/19 08:00

Diffusion-based video matting has recently emerged as a leading approach, leveraging video diffusion priors from internet-scale pretraining to achieve strong zero-shot generalization from limited synthetic data. However, the heavy compute of these diffusion backbones limits inference to a short chun

On the Fragility of Tasks under Composition: Implicit Budgeting in Language Models

👍 7

10/19 08:00

Modern language models are increasingly asked to handle several tasks within a single prompt. Doing so requires solving each task correctly while integrating the resulting answers into a coherent response. In this paper, we identify a fundamental limitation in this process, where models make substan

PVBench: Quantifying Physical Plausibility in Video Generation with Object-Level Physical Measurements

👍 5

10/19 08:00

Recent video generation models produce visually compelling outputs, yet their physical plausibility remains poorly measured. Physical plausibility asks whether generated objects obey geometry, kinematics, dynamics, and conservation laws. Existing benchmarks rely on coarse visual-language model judgm

Unsupervised Tool Evolution for Coding Agents in Data-Lake Analytics

👍 10

10/19 08:00

Coding agents can answer analytical questions over a data lake by discovering, cleaning, and joining the relevant data. However, they discard this work after each question and redo it from scratch for the next one. Tool evolution, which automatically builds and refines reusable tools across tasks, i

On the Limits of Scale: Inverse Scaling in Vision-Language Models

👍 3

10/19 08:00

Scaling laws suggest that increasing model size, training data, and compute generally yields monotonic performance improvements. While this often holds at the aggregate level, we present systematic evidence that vision-language models (VLMs) exhibit inverse scaling at the subtask level: specific que

More Memory, Worse Agents: Error Reproduction and Anti-Persistence in LLM Agents

👍 7

10/19 08:00

LLM agents can improve across tasks without updating model parameters by storing experience from their own executions in external memory and reusing it to guide later decisions. A central design question is whether to retain only observed trajectories and outcomes or also the rules, summaries, and s

Breaking Diffusion: Parallel Reasoning with Draft Refinement

👍 18

10/19 08:00

Drafting an answer allows a reasoning model to reconsider each decision in the context of the others. Yet repeated diffusion denoising does not ensure that individually plausible token updates form a better answer together. We introduce Draft Refinement Language Models (DRLMs), which use a single de

BiDex-100: A Comprehensive Benchmark for Bimanual Dexterous Manipulation Across 100+ Simulation Tasks

👍 1

10/18 08:00

Bimanual manipulation with dexterous hands has attracted growing attention in research on robotic action models. However, few studies have developed benchmarks specifically tailored for dexterous manipulation, leaving the dexterous manipulation capabilities of existing models insufficiently characte

PhysLDM: Latent Diffusion for High-Fidelity Deformable Simulation

👍 5

10/18 08:00

Neural simulation of high-fidelity deformable bodies is a foundational challenge in computer graphics and physical AI. Long-horizon prediction for high-resolution 3D volumetric meshes is hard: autoregressive methods are prone to error accumulation, while direct multi-frame prediction at native resol

Epi: Towards a Persistent, Self-Improving Agent for Autonomous Research

👍 16

10/18 08:00

We present Épi, an autonomous research agent that sustains investigations across model sessions and evolves its methods through accumulated experience. Épi combines persistent research history, hierarchical memory, branching and independent review to maintain research continuity, learn from experime

CoEvo-RSI: When Harnesses Evolve, Which Models Should Follow?

👍 59

10/18 08:00

Model--harness co-evolution has emerged as a prominent paradigm for recursive self-improving (RSI) agents, jointly evolving their execution scaffolds and intrinsic capabilities. We study the persistent yet often overlooked brain--body mismatch problem in this paradigm, where an evolving harness (bod

Axiomatic World Modeling for Physics Reasoning

👍 1

10/17 08:00

Recent research-level physics-reasoning benchmarks for LLMs reveal a failure mode that lies not in knowledge or execution, but in modeling the world being reasoned about. Models often silently alter given quantities, fabricate laws, or inject unstated assumptions to force a problem into familiar pat

Muse Glimmer 30B: An On-Device Agentic Model

👍 5

10/17 08:00

We introduce Muse Glimmer, an open-source 30B parameter causal language model with perception encoder, purpose-built for autonomous agentic tasks on consumer hardware. Muse Glimmer integrates multi-step reasoning, reliable tool use and failure recovery into a single model that runs locally on consum

Good Judgment, Bad Advice: Decision-Model Controllers Raise Agent Cost Without Improving Success

👍 11

10/17 08:00

Agent scaffolds increasingly delegate small decisions (whether to stop, verify, or change approach) to cheaper models. We ask whether a typed decision model, Jev 1.13, consulted only at deterministic decision points, makes a Claude Sonnet 5.5 terminal agent cheaper without making it worse. In a pre-

On the Tip of the Tongue: First-Token Cueing as a Behavioral Probe of Recall Failures in LLMs

👍 3

10/16 08:00

Large language models often answer factual questions incorrectly in closed-book settings, and such errors are usually attributed to missing knowledge. Yet models often encode answers they fail to produce. We show that many closed-book errors are first-token recall failures: the model commits to a wr

Event-STVG: Spatio-Temporal Video Grounding for Long-Form Event Reasoning

👍 15

10/15 08:00

Spatio-Temporal Video Grounding (STVG) aims to jointly localize a target in space and time given a natural-language query. Despite continued advances, existing STVG benchmarks remain largely centered on localizing targets through instance-specific descriptions, leaving them largely decoupled from lo

Beyond Compilation: Evaluating Faithful Natural-Language-to-Lean Statement Formalization

👍 6

10/15 08:00

Lean verifies that a generated declaration is well typed, but cannot check whether it states what the user intended. We measure how far compilation overstates faithfulness in statement autoformalization without canonical Lean targets. Our criterion requires Lean compilation and agreement of two LLM

Evaluating Agentic Capabilities Before Post-Training: An End-to-End Approach

👍 8

10/14 08:00

Agentic capabilities are increasingly important for modern LLMs, yet remain difficult to evaluate during pretraining. Existing approaches typically rely on static trajectory-derived tasks or likelihood-based proxies, which are lightweight but provide limited visibility into how base models behave in

LLM2R: From Large Language Models to 3D Reassemblers

👍 20

10/13 08:00

3D part assembly aims to reconstruct a complete shape from a set of unassembled parts by predicting a rigid SE(3) transformation for each part. Because the placement of a part is rarely determined by its own geometry, the task demands reasoning over the entire part set, and existing methods, which r

EAU: Environment-Agent-User Co-evolution for Training User-Centric Agents

👍 9

10/12 08:00

Training LLM agents to interact with users over multiple turns is challenging because the learning signal depends not only on the agent, but also on the user simulator and environment, which jointly shape the agent's experience. Simulated users can be stochastic or inconsistent, while environments m

FlowState: Execution State as Memory for Long-Horizon LLM Agents

👍 15

10/08 08:00

Long-horizon tasks require LLM agents to continually draw on information from earlier interactions. However, retaining the full history increases context costs, while compressing it risks losing details needed later, and the relevance of historical information often becomes apparent as the task prog

How To Use Decision Models for Evals

@HamelHusain · 61.6K 粉丝 · 101.6K 阅 · 1.2K 赞 · 108 转

10/21 00:54

Co-authored with Antaripa Saha. Decision models have exploded onto the scene recently, starting with TypeSafe’s Jev, followed by Perplexity’s Decider, Liquid’s d1, Cloudflare’s Clef, and OpenAI’s

中文介绍 HamelHusain与Antaripa Saha合作用Decision Models进行评估，介绍了Jev、Decider、d1、Clef等决策模型的使用方法。

Broad access through consistent standards

OpenAI News · 10/22 05:00

How OpenAI works with distribution partners to uphold consistent safeguards, support responsible access, and clarify expectations in our Usage Policies.

中文介绍 OpenAI与合作伙伴合作，通过一致的标准维护安全措施，支持负责任的访问，并明确使用政策中的期望。

NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required

Hugging Face Blog · 10/22 02:26

中文介绍 NVIDIA Kumo Relational无需训练即可在关系数据上优于监督式机器学习。

How Specializing the Vocabulary of SLMs Can Cut Inference Cost by 77% in Real Applications

Hugging Face Blog · 10/22 01:35

中文介绍 通过专业化SLMs的词汇，实际应用中推理成本可降低77%。

How AT&T is rethinking in-house legal work with OpenAI

OpenAI News · 10/22 01:00

AT&T is working with OpenAI to make legal knowledge easier to access. Learn how its LegalEdge team is rethinking intake and scaling legal judgment.

中文介绍 AT&T与OpenAI合作，使法律知识更容易获取，LegalEdge团队重新思考了法律判断的获取和扩展。

Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium.

Hugging Face Blog · 10/22 00:04

中文介绍 在AWS Trainium上对TorchNeuron进行系统级分析。

Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster

Hugging Face Blog · 10/21 22:58

中文介绍 Granite Speech 5.0可更准确、更快速地转录、翻译和归因。

Claude Mods: Flipping the Harness Inside Out

Latent Space · 10/21 21:04

Claude Mods is an early form of the Attention Interface. It flips the agent harness paradigm from the harness facing the machine, to the harness facing the human.

中文介绍 Claude Mods是早期注意力界面的形式，将代理的 harness 从面向机器转变为面向人类。

ChatGPT Ads expands to 26 new countries as over 100,000 advertisers have used the platform

OpenAI News · 10/21 21:00

From local businesses to global brands, more advertisers can now reach people as they explore, compare and decide what to buy in ChatGPT.

中文介绍 ChatGPT广告扩展至26个新国家，超过10万名广告商已使用该平台。

OpenAI models are coming to Google Cloud

OpenAI News · 10/21 20:00

OpenAI and Google Cloud are bringing frontier models and Codex workflows to Gemini’s Model Garden, available through Google Cloud Marketplace.

中文介绍 OpenAI模型即将登陆Google Cloud，将前沿模型和Codex工作流程带到Gemini的Model Garden。

Patch the Planet: global response coordination as agents uncover critical vulnerabilities

OpenAI News · 10/21 15:00

Patch the Planet connects security researchers, maintainers, and incident responders to help address critical vulnerabilities across open source.

中文介绍 Patch the Planet连接安全研究人员、维护者和事件响应者，帮助解决开源中的关键漏洞。

How we built ChatGPT’s novel ad stack for user trust and relevance

OpenAI News · 10/21 15:00

Learn how OpenAI built ChatGPT’s LLM-powered ad stack to understand conversational intent, keep ads separate from answers, and deliver relevant ads at scale.

中文介绍 了解OpenAI如何构建ChatGPT的LLM驱动的广告堆栈，以理解对话意图，将广告与答案分开，并大规模地提供相关广告。

[AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback

Latent Space · 10/21 14:54

An open source “bomb” drops from China

中文介绍 中国开源项目DeepSeek DeepInfer每天运行155T toks/集群，日收入2200万美元，利润率约80%，81天收回投资。

AI for the Fortune 2000: How Tenex Builds an AI-Native SDLC — Alex Lieberman & CJ Hess

Latent Space · 10/21 13:01

Listen now (44 mins) | Paying for outcomes, building with agents, and the last mile of enterprise AI adoption

中文介绍 听Alex Lieberman和CJ Hess讨论Tenex如何构建AI原生SDLC，以及企业AI采用的最后一英里。

Using thousands of AI agents to uncover new leads in biological data

OpenAI News · 10/21 12:50

See how OpenAI, PerturbAI, and Scripps Research used distributed AI agents to uncover promising research leads in data from millions of brain cells.

中文介绍 OpenAI、PerturbAI和Scripps Research使用分布式AI代理在数百万个脑细胞的数据中揭示新的研究线索。

Democratic governance of frontier AI, part II: Empowering cyber defenders

OpenAI News · 10/21 08:00

OpenAI outlines a blueprint for empowering cyber defenders with frontier AI, stronger safeguards, modernized cybersecurity, and more resilient public institutions.

中文介绍 OpenAI概述了一个蓝图，通过前沿AI、更强的保护、现代化的网络安全和更具有韧性的公共机构来赋予网络防御者权力。