2026-09-04

Beyond Retrieval: Progressive Latent Memory Evolution for Streaming Video Understanding

👍 3

09/03 08:00

Streaming video understanding requires multimodal large language models (MLLMs) to process continuous visual inputs and respond to user queries under strict causality and bounded memory. Existing approaches typically compress historical observations into an external memory bank and retrieve query-re

WorldReward: Reward Modeling for Camera-Conditioned World Models

👍 3

09/03 08:00

Camera-conditioned world models generate interactive videos in which commanded actions should induce the expected scene changes while appearance, geometry, and temporal dynamics remain coherent. Existing rewards assess these requirements separately: geometry-based rewards estimate trajectory executi

Random Attention: Rethinking KV Cache Eviction for Efficient Reasoning

👍 40

09/03 08:00

Large language models achieve superior performance on tasks that require extended reasoning, but long chains of thought make the KV cache a severe memory bottleneck. Existing KV cache compression methods share one paradigm: score each cached token by some estimate of how much it will matter later, a

Rethinking On-Policy Distillation of Large Language Models II: One Training Example

👍 32

09/03 08:00

On-policy distillation (OPD) combines student-generated rollouts with dense token-level supervision from a teacher. Existing work has mainly studied its algorithmic behavior, leaving the role of training data unclear. We examine this role at the data-minimal limit by training on a single query. One-

CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation

👍 9

09/03 08:00

MLLM-based embedding models remain limited in compositional retrieval, often failing to distinguish scenes containing the same concepts but different attribute-object bindings. Yet the same backbone can resolve such distinctions when used as a cross-attentive reranker, motivating us to distill its c

Editable Visual Design

👍 0

09/03 08:00

While diffusion base models such as GPT-Image-2 and Nano-Banana exhibit remarkable visual expressiveness, their end-to-end generation inherently yields flattened bitmaps with error-prone text, precluding layer-wise post-editing. Conversely, code-based visual generation via Coding Agents provides pre

Environment Evolution for Terminal Agents

👍 2

09/03 08:00

Scaling interactive and verifiable environments is critical for training terminal agents. As frontier models become more capable, environments synthesized from scratch become less challenging and thus provide limited learning signals. Recent co-evolution methods iteratively synthesize environments n

Principia: Relational Physics Tests for Video Models

👍 1

09/03 08:00

Evaluating physical reasoning in video models is difficult because absolute motion measurements depend on frame rate, object scale, and camera calibration, all of which are often ambiguous or unavailable in generated video. We propose a different approach. When two objects in the same scene obey the

Terminal-Universe: Turning Agent Trajectories into Scalable Terminal Environments

👍 6

09/03 08:00

As terminal-based code agents become prevalent, agent trajectories have accumulated at scale, while realistic, executable environments remain scarce. However, environments are what agent post-training actually requires: each can be re-queried into many verifiable tasks and provides execution feedbac

Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM

👍 47

09/03 08:00

Hybrid LLMs pair softmax attention with linear-attention layers such as Gated DeltaNet (GDN), whose recurrent state summarizes the context in fixed size. Early community 4-bit quantizations of Qwen3.8-27B (48 GDN layers, 16 attention layers) left the GDN block in 8- or 16-bit precision -- especially

Debias-SparseGPT: Bias-Aware Pruning for Large Language Models

👍 2

09/02 08:00

Model compression techniques such as pruning and quantization facilitate the efficient deployment and acceleration of Large Language Models (LLMs). However, recent studies show that weight sparsification methods, such as SparseGPT, can amplify existing biases in models, with outputs varying signific

Exploring Collaboration between a language and a non-language agent

👍 5

09/02 08:00

LLMs are increasingly deployed as orchestrators that coordinate specialized subagents to solve complex tasks through natural language. However, in many important domains like game playing and robotics, the strongest available agents are not language models. Integrating non-language agents with LLMs

VibeVoice-ASR-Streaming Technical Report

👍 10

09/02 08:00

Traditional speaker-attributed ASR systems treated ASR and speaker diarization as two separate tasks. Recently, end-to-end models such as VibeVoice-ASR have unified the two tasks within a single model. However, existing unified models still mainly support offline recognition, making it difficult to

Repo-To-Skill: Distilling GitHub Repositories Into AI4AI Skills

👍 501

09/02 08:00

Autonomous agents are beginning to carry out machine-learning (ML) research end to end. These agents combine a model backbone with a harness for planning, execution, memory, and verification, but this architecture still leaves domain-specific know-how outside the agent. We call this missing layer op

EarlyEval: Cheaper Agent Evaluation via Early Outcome Prediction

👍 112

09/02 08:00

Evaluating LLM agents is essential for guiding their development, yet it has grown prohibitively expensive: a single pass of a frontier model over an agentic benchmark can cost hundreds to thousands of dollars, a price paid repeatedly across iterative development cycles. Prior efforts, centered on b

Replacing Training with Memory: Listwise Selection for Text-to-SQL

👍 1

09/01 08:00

Modern Text-to-SQL systems often follow generate-execute-select pipelines, generating multiple candidate queries then selecting the best one. Listwise selection, by jointly comparing multiple candidates, has been widely adopted, but fine-tuning listwise selectors is costly. We thus propose a fine-tu

Sparse Readout Prism: Explaining Logit-Lens Scores in Features Instead of Tokens

👍 2

09/01 08:00

A language model's prediction of its next token develops across layers, and lens methods track this process by decoding intermediate hidden states into tokens. But a lens reading reflects both the hidden state and the readout (the unembedding matrix) used to decode it. Many lenses are fit on a corpu

Autoregressive Mosaics: Probing 2D Spatial Reasoning in Text-Only Language Models

👍 3

09/01 08:00

Large language models (LLMs) trained only on text and code can sometimes generate programs that draw recognizable images. However, it is unclear whether this reflects an internal representation of 2D spatial layout or simply the ability to translate spatial descriptions into code. We introduce Autor

ZipTok3D: High-Fidelity 3D Tokenization with Compact Token Prefixes

👍 22

09/01 08:00

Compact token sequences are essential for efficient 3D generation. However, existing 3D tokenizers typically organize latent representations either over spatial regions or as fixed-size sets of global tokens, both suffering sharp reconstruction degradation when compressed to extremely low token budg

CRISP: Cliff-awaRe Input-adaptive Sparse Prefilling with Structural-Mass-Motivated Routing

👍 8

09/01 08:00

The attention prefilling phase of long-context LLM inference scales quadratically, making self-attention a severe computational bottleneck. Traditional sparse attention methods mitigate this through fixed patterns or offline profiling, but lack the flexibility to adapt to input-dependent attention s

HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?

👍 227

09/01 08:00

As agents move from research prototypes to deployed tools, their capability increasingly depends on model-external execution infrastructure, commonly termed the agent harness. Changing this harness while holding model weights fixed can substantially alter task performance. Current agent evaluations

RealSWE: A Compositional Evaluation of Coding Agents under Realistic User Requests

👍 6

08/31 08:00

Coding agents are now commonly evaluated on the SWE-bench family of benchmarks, whose tasks are built from curated GitHub issues: long, structured, and information-rich. Real user requests, however, are typically far shorter and less structured. To characterize this gap, we define a six-category inf

WHALE: A Simple Recipe for Joint Harness-Weight Optimization

👍 28

08/31 08:00

Agent performance depends jointly on the model parameters and the executable harness code that manages context and control flow. Optimizing either component in isolation can leave the system bottlenecked by its frozen counterpart: weight updates can change which harness is effective, while harness u

FoldingAgent: Inferring Parametric Origami Procedures from Demonstration Videos

👍 3

08/31 08:00

We present FoldingAgent, an agentic framework for inferring explicit parametric folding programs directly from origami demonstration videos. Our framework leverages the reasoning power of a pre-trained Vision-Language Model (VLM) equipped with a suite of specialized tools that enable the agent to si

Ignorance or Incompetence? Constructing Knowledge-Gated, Verifiable Tasks for LLM Agents

👍 2

08/31 08:00

Professional agent tasks often depend on conventions that are absent from public corpora, yet benchmarks rarely control whether an agent has access to those conventions. We introduce a knowledge-gated task-construction protocol that separates a task instruction from a compact artefact containing pri

Aspire: Can Models Self-Evolve from Vague Goals?

👍 206

08/31 08:00

Many important forms of human learning begin with a vague goal, such as "become a better physicist" or "improve at research." Learners must interpret the goal, identify capability gaps, decide how to learn, and determine whether they have actually improved. In contrast, existing work on LLM self-evo

S3Gym: Can LLMs Turn Self-Testing and Self-Judging into Self-Improvement?

👍 27

08/31 08:00

Large language models (LLMs) increasingly interact with external environments and accumulate substantial behavioral experience, yet existing agent benchmarks largely evaluate them as fixed policies. It therefore remains unclear whether an agent can actively test its behavior, judge the resulting exp

Small Language Models as Judges for Rubric-Based Reinforcement Learning

👍 1

08/30 08:00

Rubric-based reinforcement learning extends RL beyond tasks with exact answers or rule-based verifiers by scoring responses against instance-specific criteria. However, this makes reward computation expensive: training requires repeated rubric judging, often with proprietary APIs or local generative

SnapBench: Benchmarking Snap-and-Ask Multimodal Retrieval for Mobile Interactions

👍 5

08/30 08:00

Mobile AI acts as a visual oracle, empowering users to snap a picture of something and ask for information. Snap-and-ask retrieval is now one of the most common entry points for mobile AI, yet photos are often blurry, while text questions may be short or mistyped. Existing benchmarks only test on cl

Beyond Visual Similarity: Entity-Aligned Retrieval for Knowledge-Based Visual Question Answering

👍 26

08/19 08:00

Knowledge-Based Visual Question Answering (KB-VQA) relies on retrieving external information to answer queries involving long-tail entities. However, existing retrieval pipelines predominantly employ CLIP-style dual encoders, which prioritize surface-level visual similarity over entity-level semanti

The Cognitive Revolution (When Machines Do The Thinking)

@Konstantine · 17.6K 粉丝 · 256.2K 阅 · 529 赞 · 100 转

09/02 23:07

We externalized our muscles and built the modern world. Now we are externalizing our minds. Most mornings I ride to work in a Waymo. The car is doing physical work: five thousand pounds of metal and

中文介绍 Konstantine探讨认知革命，从肌肉外化到思维外化，引出AI在现代社会中的角色。

Nobody is talking seriously about AI demand

@giovannicatt3 · 1.4K 粉丝 · 142.0K 阅 · 507 赞 · 56 转

09/01 23:51

AI capex for 2028 is forecast to be larger than the budget of France. Frontier AI labs’ revenue ramp justifies almost any number. Reflexivity in AI demand is a double-edged sword, and it's now a good

中文介绍 giovannicatt3分析2028年AI资本支出，预测将超过法国预算，并探讨AI需求的双刃剑效应。

Prediction: AI will collapse

@wordgrammer · 26.1K 粉丝 · 120.8K 阅 · 602 赞 · 33 转

09/03 05:30

The dot com bubble got some things right. It noticed that the internet was going to be big. It noticed that the value generated by the internet would roughly scale in proportion with the amount of

中文介绍 wordgrammer预测AI将面临崩溃，类比互联网泡沫，指出互联网价值与规模的比例关系。

How to design a Brand System from 1 reference image

@AmirMushich · 80.6K 粉丝 · 102.2K 阅 · 501 赞 · 42 转

09/03 00:11

For 10+ years, I've been creating branding & ads for Warner Music, PepsiCo, and others. Now I combine that experience with AI to build hi-end visual systems faster. This article shows my complete

中文介绍 AmirMushich分享从单一参考图像设计品牌系统的经验，结合AI快速构建高端视觉系统。

Compilers 2.0: AI as stochastic optimizer

@cdleary · 2.3K 粉丝 · 100.3K 阅 · 533 赞 · 66 转

09/02 04:00

There has been a lot of discussion following the presentation of the Jalapeño MLA kernel at HotChips and subsequent commentary by SemiAnalysis. As OpenAI’s hardware team, we just barely touched on

中文介绍 cdleary介绍Jalapeño MLA内核，讨论AI作为随机优化器在编译器中的应用。

7 Grok Bots for Marketing

@irabukht · 18.4K 粉丝 · 63.3K 阅 · 536 赞 · 28 转

09/02 07:39

18 things to know and 7 paste-in prompts for running ads, SEO and GEO on Grok Bot. Collected from 1,000+ marketers in our community. Most of us use Grok Bot to run marketing work end to end: Weekly

中文介绍 irabukht推荐7款Grok Bot营销工具，分享18个要点和7个粘贴提示，用于广告、SEO和GEO营销。

Introducing agentic video understanding with Gemini

@GoogleAIStudio · 198.5K 粉丝 · 37.4K 阅 · 522 赞 · 48 转

09/02 02:02

Our new agentic feature for video analysis cuts costs by up to 66% and reduces token consumption by up to 88% while boosting accuracy. Today, we’re launching agentic video understanding across our

中文介绍 GoogleAIStudio推出Gemini视频分析新功能，降低成本并提高准确性。

Using Grok Bot: 8 templates to get inspired

@mattyp · 57.4K 粉丝 · 29.7K 阅 · 508 赞 · 40 转

09/02 01:03

Templates are now available for Grok Bot, so I wanted to share my favorites. Templates contain the skills, memories, and official @bot plugins from your bots. When you use a template, you get a copy

中文介绍 mattyp分享Grok Bot的8个模板，包含技能、记忆和官方插件，以激发灵感。

Introducing Gemini 3.8 Flash and 3.8 Flash Cyber

@GoogleAIStudio · 198.5K 粉丝 · 26.2K 阅 · 507 赞 · 59 转

09/03 00:48

Our newest Gemini models deliver next-generation intelligence for agentic workflows and cybersecurity. Building on the momentum of 3.7 Flash from three weeks ago and marking our third Flash release in

中文介绍 GoogleAIStudio发布Gemini 3.8 Flash和3.8 Flash Cyber模型，为工作流和网络安全提供下一代智能。

From rg to zg: Local Search Beyond Keywords

@QwenDevs · 10.9K 粉丝 · 21.4K 阅 · 509 赞 · 55 转

09/02 22:30

Summary: The information humans and agents need is often scattered across large numbers of local files, making it difficult to locate accurately and efficiently. zg (zvec-grep) is local-first search

中文介绍 QwenDevs介绍zg（zvec-grep）本地搜索工具，帮助快速准确定位大量本地文件中的信息。

GPT-6 Astra: an automated AI Engineer you can hire for <$6 an hour

Latent Space · 09/04 05:09

We spent 20B+ tokens of GPT-6 Astra to explore everything. Here’s our learnings.

中文介绍 Latent Space推出GPT-6 Astra，每小时费用低于6美元，用于自动化AI工程。

Introducing WeatherNext 3, our most advanced and accurate global weather AI model

DeepMind Blog · 09/03 23:02

中文介绍 DeepMind推出最先进和准确的全球天气AI模型WeatherNext 3。

Daybreak for Frontline Defenders: $1B to protect essential services

OpenAI News · 09/03 21:15

OpenAI introduces Daybreak for Frontline Defenders. A $1 billion commitment expands access to frontier cyber AI, training, and support for essential services.

中文介绍 OpenAI推出Daybreak项目，投入10亿美元用于保护关键服务，扩大前沿网络AI的培训和支持。

NeoMME: an efficient Multimodal-native and Multilingual Encoder

Hugging Face Blog · 09/03 21:13

中文介绍 Hugging Face发布NeoMME，一个高效的多模态和多语言编码器。

Legora reviewed 41 documents in minutes with GPT-6 Astra

OpenAI News · 09/03 20:00

Legora used GPT-6 Astra to review 41 documents in minutes, find all four planted errors, and improve performance by nearly 40% in this financial-review workflow.

中文介绍 Legora使用GPT-6 Astra在几分钟内审查41份文件，发现所有四个植入的错误，并在此财务审查流程中提高性能近40%。

Playco cut manual fixes 50% prototyping games with GPT-6 Astra

OpenAI News · 09/03 20:00

Using GPT-6 Astra, Playco built three themed game prototypes from one grey box foundation and reported 50% fewer manual fixes than with the previous model.

中文介绍 Playco使用GPT-6 Astra减少50%的手动修复，构建了三个主题游戏原型。

[AINews] Muse Spark 1.3 matches GPT-5.6-Sol, confirming Meta Superintelligence as the newest Frontier Lab, >90% discount for training

Latent Space · 09/03 12:38

an epic comeback story for Meta

中文介绍 Meta的Muse Spark 1.3与GPT-5.6-Sol匹配，确认Meta超级智能为最新前沿实验室，训练折扣超过90%。

Safety overview: GPT-6 Astra

OpenAI News · 09/03 08:00

GPT-6 Astra is our most capable broadly deployed model and our first to reach the Critical level of cybersecurity capability under our Preparedness Framework.

中文介绍 GPT-6 Astra是OpenAI最强大、广泛部署的模型，首次达到我们准备框架下的关键网络安全能力水平。

Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps

Hugging Face Blog · 09/03 08:00

中文介绍 Hugging Face在100个GRPO步骤中微调350M模型，以获得更好的结构化输出。

Give Your Coding Agents a Memory You Own

Hugging Face Blog · 09/03 08:00

中文介绍 Hugging Face提出为编码代理提供自有记忆的方法。

Training a coding model to paint watercolours with TRL and OpenEnv

Hugging Face Blog · 09/03 08:00

中文介绍 Hugging Face使用TRL和OpenEnv训练编码模型以绘画水彩画。

Muse Spark 1.3 ✨, Gemini 3.8 Flash ⚡, intelligence vs cost 📊

TLDR AI · 09/03 08:00

中文介绍 TLDR AI报道Muse Spark 1.3和Gemini 3.8 Flash，讨论智能与成本的关系。

Proactive cyber defense for governments and enterprises

DeepMind Blog · 09/03 00:24

中文介绍 DeepMind为政府和企业提供主动网络防御。

Introducing Gemini 3.8 Flash and 3.8 Flash Cyber

DeepMind Blog · 09/03 00:18

中文介绍 DeepMind推出Gemini 3.8 Flash和3.8 Flash Cyber。

Facilitating AI integration with simplicity at scale

MIT Tech Review AI · 09/02 22:00

As companies scale, the technology supporting operations can become a liability just as quickly as it becomes an asset. Disconnected systems, site-specific tools, spreadsheets, and manual workarounds can create data silos that make it harder to spot problems early, coordinate responses, and make dec

中文介绍 MIT Tech Review AI讨论如何以规模化的简单性促进AI集成。