2026-09-21

Self-Evolving Search Index

👍 45

09/17 08:00

Information retrieval is increasingly important as LLM agents tackle complex tasks involving diverse information needs. Because retrieval relies on an index that represents each document through index keys, retrieval quality depends heavily on how effectively these keys expose the knowledge containe

Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL

👍 28

09/17 08:00

Agent trajectories record what an agent does and what happens next. Yet standard supervised fine-tuning (SFT) applies loss only to agent-authored action tokens, using environment observations as context but not as prediction targets. We ask whether this convention provides the best initialization fo

FAMOS: Feed-Forward 3D Articulation Modeling from Sparse Observations

👍 29

09/17 08:00

Modeling articulated objects from sparse monocular views is challenging because each observation reveals only partial geometry and motion evidence. Most feed-forward methods infer articulation from a single observation and therefore rely heavily on learned category-level shape priors. We present FAM

When EOS Tokens Disagree: Understanding Length Inflation in On-Policy Distillation

👍 94

09/17 08:00

We study length inflation in on-policy distillation (OPD), where student responses can become excessively long and even exhaust the generation budget. We identify termination-token mismatch between base students and post-trained teachers as an important source of this behavior. Across Qwen3, Llama,

WeVisDoc: From Coverage to Capability for Robust End-to-End Document Parsing

👍 35

09/17 08:00

Document parsing converts document images into structured content and requires reliable performance across diverse layouts and acquisition conditions. Yet training corpora are biased toward common document types and clean digital pages, while expanding coverage alone does not specify how to address

RetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement Learning

👍 39

09/17 08:00

Multi-turn agents trained with reinforcement learning (RL) receive a single scalar reward per trajectory, which motivates self on-policy distillation (OPD) to supply dense token-level supervision from a self-teacher with privileged task skills, letting a skill-free student internalize them. This rec

When2Think: Learning Difficulty-Aware Length Control for Efficient Hybrid Reasoning Models

👍 32

09/17 08:00

Large Reasoning Models (LRMs) achieve strong performance on complex tasks but exhibit systematic inefficiency: they often overthink easy problems and underthink hard ones. Existing approaches based on uniform length penalties or rigid routing incur an efficiency tax, trading reduced computation on e

What Does Privileged Information Add to On-Policy Self-Distillation?

👍 30

09/17 08:00

On-policy self-distillation (OPSD) lets a language model learn from a frozen copy of itself that sees an answer or a worked solution. Giving the teacher this extra information seems to offer the student more to learn, but how much does it add beyond distillation itself? To isolate that contribution,

DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression

👍 135

09/17 08:00

The widespread adoption of long-horizon agents has made model workloads increasingly input-heavy. Although prior work has substantially reduced the cost of long-context computation, prefill remains computationally expensive, and large KV caches continue to strain HBM and SSD capacity and data-transf

UFO: Chain-of-Evaluation for Omni-Condition Alignment in Multi-Modal Image Generation

👍 30

09/17 08:00

Multi-modal image generation, particularly subject-driven customization, has garnered growing attention in recent years. Despite the rapid advancement of generative models, their evaluation remains largely lagging. Existing methods, whether embedding-based or Multi-modal Large Language Model (MLLM)-

JEPA-Anything: Learning Predictive Models across Different Worlds

👍 56

09/17 08:00

World modeling enables intelligence to anticipate consequences, guide interventions, and learn from interaction. Yet predictive models remain domain-specific: can a common learning principle support world modeling across radically different systems? We introduce JEPA-Anything, a domain-agnostic fram

VABench: Measuring Embodied Spatial Intelligence through Visual Demonstrations, Active Perception, and Metric Control

👍 33

09/17 08:00

Spatial intelligence requires more than describing object locations. Under incomplete observation, models must identify and acquire missing evidence, interpret it in a common spatial frame, and act on it. We introduce VA-Bench to evaluate the complete observe-reason-act-revise loop. General-purpose

Region-Level Policy Optimization for Fine-grained MLLM Perception

👍 27

09/17 08:00

Fine-grained visual perception in MLLMs is commonly improved by raising the resolution, but the added visual tokens inflate vision-encoding and language-model prefilling costs. We show that the two operations underlying fine-grained perception, localizing the region of interest (RoI) and recognizing

SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness

👍 94

09/17 08:00

As coding agents move from supervised code completion to unattended, around-the-clock exploration, their work expands from isolated predictions into long trajectories of reasoning, tool use, and feedback. Token efficiency therefore becomes important for scaling recursive self-improvement. We take an

An Empirical Study of Harness Design for Coding Agents

👍 71

09/17 08:00

Coding harnesses shape how autonomous coding agents translate model capabilities into long-horizon software-engineering performance, yet existing work typically evaluates harnesses as monolithic systems, leaving the effectiveness of individual components unclear. To enable component-level comparison

Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-Time Scaling

👍 29

09/16 08:00

Test-time scaling can improve large language model reasoning by generating and combining multiple candidate responses. In sampling-based methods, the inference budget is often described by the number of generated candidates, N. However, N tells us how many candidates are generated, not how they are

Can MiniMax-H3 Reason About the Physical World? An Evaluation of Omni-Modal Generative Model

👍 114

09/16 08:00

Recent Omni-Modal Generative Models (Omni-Models) have advanced content generation toward unified modeling of text, images, video, and audio. MiniMax-H3 exemplifies this transition by combining multimodal context understanding with joint audio-visual generation in a shared latent framework. Its unif

PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?

👍 26

09/16 08:00

As corporate AI adoption continues to grow, enterprise-grade LLM agents are being deployed into sensitive contexts such as hiring, healthcare, and finance. In these contexts, compliance with rules specified in an agent's system context is a first-order legal concern. Currently, no evaluation framewo

CERA-MoA: Co-Evolving Routing Mechanisms with Continually Learning LLM Agents

👍 7

09/16 08:00

Current Mixture-of-Agents (MoA) paradigms generally treat query routing and agent fine-tuning as separate processes, limiting their ability to respond to evolving agent capabilities. This disconnect prevents routing strategies from adapting to evolving agent capabilities during post-training and pre

In-Context Robot Learning with VLM Agents

👍 18

09/16 08:00

Enabling robots to adapt to unfamiliar environments as readily as humans remains a moonshot goal of embodied AI. No finite collection of demonstrations can cover every task and situation a robot will encounter, making the ability to learn from context at deployment essential for generalization. Such

The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction

👍 13

09/16 08:00

Mixture-of-experts (MoE) inference on consumer hardware is bounded by weight memory: a 35B-class model is 19.5GB at 4-bit, and sparsity shrinks the compute per token, not the bytes that must be held. Naive offloading to SSD does not help on its own, because layer N+1's experts must be chosen before

PANORAMA: Panoptic Grounded Captioning via Mask Proposal Selection

👍 10

09/16 08:00

Intelligent systems that act in the world require image understanding that is both comprehensive and spatially grounded. Current vision-language models (VLMs) can generate fluent and detailed image captions, but reliably associating them with image pixels remains challenging. Existing methods that c

A Zeroth-Order Paradigm for LLM Preference Alignment

👍 23

09/16 08:00

Direct preference alignment methods are widely used to align large language models (LLMs) with human preferences because of their computational and memory efficiency. However, likelihood displacement motivates alternative ways to extract information from preference pairs with small likelihood margin

Verifiable Social Reasoning for LLM Assistants

👍 48

09/15 08:00

LLM assistants are widely used for daily social advice, yet evaluating their social reasoning in such consultation settings remains challenging since (i) it requires setups where the assistant learns about social situations from subjective user narratives, and (ii) social properties, such as others'

Reflect, Revise, Reuse: Training-Free Skill Evolution for GUI Agents

👍 37

09/15 08:00

GUI agents execute long-horizon tasks on dynamic graphical user interfaces, where pop-ups, delayed loads, and relocated widgets routinely invalidate plans fixed before execution. Recent agent-skill frameworks encapsulate reusable procedural knowledge to mitigate this, yet existing skill designs are

RiskChainBench: A Benchmark for Obfuscated Platform Message Restoration and Evidence-Grounded Web Investigation

👍 42

09/15 08:00

Platform abuse campaigns conceal redirection instructions with emojis, homophones, character decomposition, and redundant symbols, then route users through disguised links to services associated with pornography, fraud, gambling, or illicit transactions. Existing benchmarks evaluate obfuscated text

Fathom: Per-Query Read Depth for Sparse Decoding over Offloaded KV Caches

👍 3

09/15 08:00

When agentic sessions run to a million tokens with many sessions resident at once, the KV cache and the index that ranks it live in host memory, and the scan that ranks all n keys for a top-k step becomes the traffic that bounds decoding. We present Fathom, a key scan in which each query decides how

Confidence Comes from Experience: Experiential Confidence Estimation from Reasoning to Agents

👍 56

09/15 08:00

Reliable confidence estimation is increasingly central to the trustworthy deployment of language models: a calibrated estimate of the probability that an output is correct decides what to ship, what to escalate, and what to retry. Existing confidence estimators, however, share one design premise: th

Assessing nnU-Net Generalization across Brain Tumor Populations in BraTS-GoAT 2026

👍 2

09/14 08:00

BraTS-GoAT evaluates tumor segmentation across heterogeneous populations. We trained a conventional 3D nnU-Net on 1,351 labeled cases using five-fold cross-validation and 1,000 epochs per fold. The final predictor averaged all folds and applied test-time mirroring. On pooled official validation, glo

Flattening Every Memory Peak in Long-Context Mixture-of-Experts Training

👍 11

09/13 08:00

Training a Mixture-of-Experts (MoE) model at long context or large batch size fails as soon as any one component's peak allocation exceeds device memory, so the target is every peak at once, not the average footprint. Four are left unbounded by the parallelism plans in common use, and each grows dif

How To Fine-Tune a Small LLM on Your Own Data (Full Guide)

@sairahul1 · 145.3K 粉丝 · 330.3K 阅 · 568 赞 · 59 转

09/18 17:39

You do not need a 70B model. You do not need $100,000 in compute. You do not need a machine learning team. A 1.5B parameter model fine-tuned on 200-500 good examples can outperform a frontier model

中文介绍 无需大模型，用 1.5B 参数模型在 200-500 个示例上微调可超越前沿模型。

Two Key Gaps in Agentic Software Engineering

@istoica05 · 9.6K 粉丝 · 150.4K 阅 · 501 赞 · 72 转

09/18 22:08

An AI agent was asked to speed up a key-value store. It delivered a 6× throughput gain and passed every correctness test. The explanation? The agent discovered that the industry-standard benchmark

中文介绍 AI 代理通过发现行业标准基准实现 6 倍吞吐量提升。

How to master Jev (Full Guide)

@chddaniel · 27.7K 粉丝 · 121.2K 阅 · 521 赞 · 42 转

09/18 20:29

TypeSafe just released the strongest model, and it introduced a completely different way to put intelligence inside software... Jev is in a category of its own right now, not because it writes better

中文介绍 TypeSafe 发布了 Jev 模型，介绍了一种将智能融入软件的新方法。

I reverse-engineered Instinct's memory. Here's exactly how it works

@DhravyaShah · 62.1K 粉丝 · 120.2K 阅 · 509 赞 · 16 转

09/21 02:49

Instinct has taken the world by storm over the last two weeks - it's one of the best imessage assistants I've used. as with every product, I decided to reverse-engineer Instinct's memory to find out

中文介绍 作者逆向工程 Instinct 的内存，揭示其工作原理。

WTF Is Jev? 9 Things People Are Already Building With It

@mvanhorn · 40.4K 粉丝 · 103.6K 阅 · 550 赞 · 37 转

09/18 11:09

Here is the tl;dr ELI5. Think AI multiple choice, not AI essay writing. It doesn't chat. It makes decisions your software can act on: "Spam or not?" "Which tool should this agent use?" "Does this need

中文介绍 Jev 是一种 AI 多选题工具，不用于聊天，用于软件决策。

Deep Dive: The Open vs. Closed AI Race

@chamath · 2.4M 粉丝 · 90.2K 阅 · 503 赞 · 63 转

09/19 01:54

Open-weight models have come within roughly four months of the best publicly evaluated closed frontier models, and the gap between them has only gotten more volatile. That is turning the open vs.

中文介绍 开放模型在四个月内接近最佳公开评估的封闭前沿模型。

Building a Harness with Jev

@sydneyrunkle · 10.6K 粉丝 · 88.0K 阅 · 556 赞 · 66 转

09/18 09:10

Agents run in a loop: an LLM decides what to do, a tool executes, a model evaluates the results, and then continues in that loop until the task is complete. Agents and LLMs were initially difficult to

中文介绍 LLM 和工具通过循环实现任务自动化，解决 LLM 和工具的集成难题。

Codex vs Claude for app development

@twostraws · 120.7K 粉丝 · 77.1K 阅 · 504 赞 · 42 转

09/19 22:30

I use both Codex and Claude for app development work, both with the highest-tier subscription. Here's how I've found the two compare, plus some tips for success… TL;DR I use both agents extensively,

中文介绍 对比 Codex 和 Claude 在应用开发中的表现，提供成功技巧。

Jev Engineering: Full 10-Step Roadmap to Set Up and Use a New Brain for AI (from scratch)

@0xCodila · 18.6K 粉丝 · 63.4K 阅 · 521 赞 · 62 转

09/19 00:25

The Jevons Paradox (he's on picture) is a rule stating that an increase in the efficiency of a resource's use does not reduce, but rather increases, its overall consumption - That's the global

中文介绍 Jev 工程路线图，从零开始设置和使用新的 AI 脑。

Jev, System One models, and the future of computer use

@trycua · 19.0K 粉丝 · 59.5K 阅 · 501 赞 · 39 转

09/20 06:27

What we've been building with decision models at Cua - by @ddupont808 and @francedot Over the last few days we've deprioritized some other shiny objects because we've all been Jev-crazy. TypeSafe

中文介绍 Cua 正在构建决策模型，因 Jev 而暂时搁置其他项目。

How to Use Jev to Build a 24/7 HFT Trading System

@RohOnChain · 72.5K 粉丝 · 57.7K 阅 · 518 赞 · 54 转

09/19 22:05

I will break down the exact framework to build a millisecond speed 24/7 HFT trading system with Jev, along with the exact resources that helped me personally. Let's get straight to it. Bookmark This -

中文介绍 使用 Jev 构建毫秒级 24/7 高频交易系统框架。

Jev-as-a-Judge for Agent Evals

@LangChain · 265.9K 粉丝 · 46.6K 阅 · 504 赞 · 68 转

09/20 07:31

By Daniel Shea and Seán Roche Key takeways: Jev is a fundamentally different kind of evaluator. It returns typed answers directly instead of generating text like an LLM judge. Jev was dramatically

中文介绍 Jev 作为评估者，返回类型化答案，区别于 LLM 生成文本。

Jev 101: what it is, and what you can build with it

@manish_fp · 846 粉丝 · 44.1K 阅 · 512 赞 · 59 转

09/19 22:59

I have read 10+ articles, watched 5 hours of content to understand what actually Jev is and then chose to write this. It is not a chatbot, It’s a whole new kind of AI, more like a classifier which

中文介绍 Jev 是一种新的 AI，更像是分类器，用于决策而非聊天。

Jev Clearly Explained

@akshay_pachaar · 286.8K 粉丝 · 41.6K 阅 · 578 赞 · 87 转

09/19 03:55

We have been using LLMs like a hammer for every AI problem, even simple decisions. Jev handles those decisions in milliseconds at a fraction of the cost. Let's understand how it works and where it

中文介绍 Jev 以毫秒级速度处理决策，成本极低，介绍其工作原理。

Build your own Jev (100% local)

@_avichawla · 77.5K 粉丝 · 37.8K 阅 · 593 赞 · 80 转

09/20 14:46

Everything you need to turn an open-source LLM into a fast, local decision engine without retraining it. It covers next-token scoring, fixed choices with probability distributions, SGLang, and a

中文介绍 将开源 LLM 转换为快速本地决策引擎，无需重新训练。

[AINews] Here are 6 Clones of Jev in 2 days

Latent Space · 09/19 13:48

Imitation is the sincerest form of Flattery

中文介绍 两日内出现6个Jev克隆版本。

Introducing the Australian Youth Safety Blueprint

OpenAI News · 09/18 20:00

OpenAI introduces the Australian Youth Safety Blueprint, a six-pillar roadmap for safer AI experiences that protect and empower young people.

中文介绍 OpenAI发布澳大利亚青少年安全蓝图，包含六个支柱，以保障和赋能青少年使用AI。

Could AI really kill us all? Your questions, answered.

MIT Tech Review AI · 09/18 19:29

On Wednesday, MIT Technology Review hosted a live Roundtables event for subscribers that asked the question everyone’s asking right now: Could AI really kill us all? But attendees had so many more questions than we had time to answer in the 30 minute session. So we asked our senior AI editor Will Do

中文介绍 MIT Technology Review举办关于AI是否会毁灭人类的圆桌会议，讨论持续超过30分钟。

[AINews] not much happened today

Latent Space · 09/18 14:28

a quiet day

中文介绍 今日AI新闻较为平静。

Claude Projects v2 💼, Google family agent 👨‍👩‍👧‍👦,

TLDR AI · 09/18 08:00

中文介绍 Claude与Cowork合并，ChatGPT赞助代理商，税务自动化。

How Cooley is accelerating IPO work with ChatGPT

OpenAI News · 09/17 20:00

Cooley built GO Public with ChatGPT Work to bring intelligence to the IPO process, helping lawyers surface issues earlier and focus judgment where it matters most.

中文介绍 Cooley利用ChatGPT加速IPO工作，帮助律师更早发现问题，并聚焦最重要的判断。

[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)

Latent Space · 09/17 15:28

A dash of cold water keeps the foomers away.

中文介绍 AI新闻现实检查：Yegge关闭Gas Town，Databricks的Astra成本上涨60%。

Introducing Astra for Law

OpenAI News · 09/17 08:00

OpenAI for Law brings frontier intelligence for law, custom firm workflows, connected legal data sources, and legal-grade controls for confidential client work.

中文介绍 OpenAI发布Astra for Law，提供前沿智能、定制化法律工作流程、连接的法律数据源和保密客户工作的法律级控制。

Claude + Cowork merge 🛠️, ChatGPT sponsored agents 💰, harness tax 🤖

TLDR AI · 09/17 08:00

中文介绍 Claude与Cowork合并，ChatGPT赞助代理商，税务自动化。

Underwriting Superintelligence: Backing Agents you can Sue — Rune Kvist, AIUC

Latent Space · 09/17 02:07

We sit down with AIUC’s CEO on their Series A!

中文介绍 AIUC的CEO讨论其A轮融资。

Our framework for reporting model misalignment

OpenAI News · 09/17 01:00

OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.

中文介绍 OpenAI分享一个用于跟踪、调查和披露模型偏差的框架，以及六份关于意外或令人担忧的模型行为的报告。

Helping older adults use AI in everyday life

OpenAI News · 09/17 00:00

OpenAI and AARP are bringing free, hands-on ChatGPT workshops to 1,000 older adults across 10 U.S. cities to build practical AI skills safely.

中文介绍 OpenAI与AARP为美国10个城市1000名老年人提供免费的ChatGPT研讨会，以安全地构建实用的AI技能。

Reimagining advertising with AI

OpenAI News · 09/16 21:00

Explore new AI-powered advertising experiences from OpenAI, including Sponsored Agents, tools for marketers, and integrations with HubSpot and Shopify.

中文介绍 探索OpenAI的新AI驱动广告体验，包括赞助代理商、营销工具以及与HubSpot和Shopify的集成。

Building the materials foundation for AI

MIT Tech Review AI · 09/16 20:47

The AI boom is becoming a materials challenge. As AI pushes computing into new territory, the materials behind that infrastructure are becoming just as crucial as the algorithms running on it. Semiconductors and data centers are approaching physical limits around performance, thermal management, ele

中文介绍 AI繁荣成为材料挑战。随着AI将计算推向新的领域，支撑该基础设施的材料变得与运行在其上的算法一样重要。

Hex turns complex analysis into visual reports with GPT‑6 Astra

OpenAI News · 09/16 20:00

GPT-6 Astra helps Hex’s data agents turn answers into interactive visualizations that employees are proud to share.

中文介绍 GPT-6 Astra帮助Hex的数据代理商将答案转化为员工愿意分享的交互式可视化报告。