2026-09-18

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

👍 4

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

👍 14

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

👍 8

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

👍 9

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

👍 22

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

Agora: Git as Shared Memory for Collective AutoResearch

👍 39

09/16 08:00

Autonomous research loops such as AutoResearch show that one coding agent can improve a training setup unattended. Run several of them and each session starts from scratch, so more agents tend to mean more duplicated search rather than more discovery. Agora is a shared memory for such agents: resear

ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments

👍 70

09/16 08:00

Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience-a challenge we call the sci

ProgramDistill: From Interactive Web Apps to Verifiable Reference-Guided SWE Tasks

👍 44

09/16 08:00

Coding agents are typically evaluated with desired behavior specified through issues or instructions. In practical web development, however, agents may need to infer behavior from working software and implement it in an incomplete application. We introduce ProgramDistill, a benchmark evaluating codi

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

👍 2

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

👍 48

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

EvolveTrade: Experience-Driven Policy Refinement for Self-Evolving LLM Trading Agents

👍 30

09/15 08:00

Large language model (LLM) trading agents can combine market data, news, and executable analysis, but their behavior is often controlled by static hand-written tool-use policies that are fixed before deployment. This limits their ability to adapt how they gather evidence, invoke tools, verify signal

ImpossibleRubrics: Stress-Testing Generated Rubrics as Reward Signals

👍 3

09/15 08:00

Language model-generated rubrics are increasingly used as reward signals for rubric-based reinforcement learning, LLM-as-a-judge evaluation, and automated grading. Such rubrics are reliable only if they reward honest answers over adversarial answers optimized to exploit them. Yet their robustness to

Emergence World: Adversarial Stress-Testing of Long-Horizon Multi-Agent Systems

👍 2

09/15 08:00

As AI agents move from bounded tasks to persistent deployments, failures can propagate through memory, tools, other agents, and environmental state long after their interactions. This creates a safety regime that cannot be characterized by evaluating model responses in isolation. Emergence World, is

Modality-Autoregressive World-Action Models

👍 7

09/15 08:00

World-action models (WAMs) jointly model future observations and actions, typically predicting the future as RGB images. Other visual modalities such as depth, pretrained visual features, and point tracks can more efficiently capture geometric, semantic, and motion features. However, how best to com

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

HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses

👍 25

09/14 08:00

Scientific agents contribute to hypothesis discovery by synthesizing evidence, assessing proposals, and developing new explanations. Recent systems combine scientific agents with evolutionary search through critique, comparison, and revision. However, how different forms of agent collaboration affec

Register Tokens for Bounded-State Reasoning in Diffusion Language Models

👍 4

09/14 08:00

Masked diffusion language models (dLLMs) generate text by iteratively denoising masked tokens with bidirectional attention. Extending reasoning across generation chunks normally requires keeping earlier generated text in context. We ask whether a dLLM can instead continue reasoning after that text i

ModularRSI: Modular and Generalizable Recursive Harness Self-Improvement

👍 10

09/14 08:00

Recent work extends recursive self-improvement (RSI) to agent harnesses for long-horizon coding and terminal tasks, enabling agents to improve execution mechanisms from experience. However, generalizable harness RSI remains challenging. First, evolving harnesses on evaluation benchmarks or their sub

HarnessVLN: Unifying Training-Free Embodied Navigation through an Agent Harness

👍 17

09/14 08:00

Embodied navigation requires agents to interpret visual observations, accumulate spatial knowledge, and execute actions to follow instructions or locate objects. Training-based methods face generalization challenges, while training-free methods exploit multimodal large language models (MLLMs) but of

Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration

👍 2

09/14 08:00

Safety guardrails in open-weight language models can be readily bypassed using Refusal Feature Ablation (RFA), a technique that identifies and projects out a linear refusal direction from the residual stream, often achieving a high attack success rate (ASR) while preserving model capability. Defendi

Mind2Dialogue: Training Human-Aware Language Models by Simulating User Mental States

👍 5

09/14 08:00

As language models become more capable, long-term collaboration in learning, reasoning, and decision-making calls for a deeper understanding of the people they serve. Yet training such human-aware language models faces a fundamental supervision gap because current datasets for LLM assistant training

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

👍 9

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

OmniHarness: Harnessing Generalizable Visual Generation via Symbolic Policy Learning

👍 2

09/13 08:00

Unified multimodal large language models (MLLMs) and multi-agent systems have advanced visual generation. However, three limitations remain. (1) Existing methods often distill task-specific experience with limited generalizability. (2) Reflection is often deferred until task completion. (3) Knowledg

Another Blueprint In The Wall: How to Ask Frontier AI Like a Kid?

👍 7

09/13 08:00

This paper reports experiments across six frontier model types from OpenAI, Anthropic, xAI, and Google DeepMind. Ten independent sessions per model type used the same three stage prompt sequence, progressing from architectural preference to a full ASCII backbone. Under the school audience framing, r

StepAudio 3 Realtime Technical Report

👍 102

09/12 08:00

Realtime spoken interaction demands deep reasoning, prompt responses, and fluid turn-taking. We present StepAudio 3 Realtime, an audio-language foundation model organized around a continuous listen-converse-think-act loop. Deep Perception captures rich acoustic cues to interpret user intent, while S

Drift-Constrained Optimization: Only Direction Matters in Fine-Tuning Instruct Models

👍 2

09/12 08:00

Fine-tuning instruct models often improves target performance while inducing behavioral drift from the reference model, which can degrade existing capabilities. Rather than treating this drift as an uncontrolled consequence of optimization, we specify a behavioral drift budget before optimization an

Training Specialist Models without Reasoning Trajectories for Domain Expert Distillation

👍 3

09/12 08:00

Specialist distillation effectively transfers domain expertise to student models via teacher-generated reasoning trajectories. However, when these specialists are trained solely on question--answer pairs without explicit reasoning supervision, what governs the trajectories they generate? In this wor

Generalized Agent Iteration: One Formal Framework for Iterative Policy Improvement and Recursive Self-Improvement

👍 59

09/11 08:00

When we speak of recursive self-improvement (RSI), are we speaking of a phenomenon, a mechanism, or a prospect? Towards autonomous and evolving intelligence, RSI is being claimed at many scales, while no single framework that formally describes these emerging instances exists. Its counterpart in the

StepAudio 3 Music Technical Report

👍 76

09/11 08:00

We introduce StepAudio 3 Music, a large-scale, long-form music generation model that supports explicit musical planning and open-domain text-controlled generation. The StepAudio Music Tokenizer represents audio as a 50-Hz stream from a 65536-entry single codebook, using semantically informed self-su

The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement

👍 88

09/10 08:00

Recursive self-improvement (RSI) enables AI systems to turn experience and feedback into persistent changes that improve both their capabilities and the process of future improvement. We first use the Headroom-Closed Index (HCI) to reveal the problems of existing LLMs, then introduce the RSI concept

Its time to talk about ‘model welfare’

@mustafasuleyman · 1.1M 粉丝 · 204.3K 阅 · 502 赞 · 86 转

09/16 22:01

AIs are not conscious. They do not feel, experience, or suffer. However, there’s a growing movement in support of model welfare; the idea that we might soon owe them a duty of care. I think this

中文介绍 探讨“模型福利”，即人工智能模型应获得关怀的观点。

THE ARCHITECTURE OF AI-NATIVE WEB3.0

@LingoAITech · 411.4K 粉丝 · 156.5K 阅 · 524 赞 · 227 转

09/15 23:43

From the Semantic Web to the Human-Centric Agentic Economy Data3.0 + Finance3.0: Sovereign Data, Personal AI, Open Agents and Programmable Value Abstract Artificial intelligence is changing the role

中文介绍 分析AI原生Web3.0架构，从语义网到以人为中心的智能经济。

open the frontier

@jack · 12.1M 粉丝 · 153.5K 阅 · 2.2K 赞 · 304 转

09/15 07:59

the frontier is the edge of what we know. no company owns what comes next. i want more people to be able to advance it. i favor open releases that people can examine, use, and improve together without

中文介绍 提倡开源AI前沿技术，鼓励共同探讨、使用和改进。

Adopting the software factory model: crawl, walk, run

@zachlloydtweets · 20.4K 粉丝 · 123.1K 阅 · 500 赞 · 47 转

09/16 03:19

The software factory approach (a closed agentic loop that runs in the cloud) is growing in popularity, but it can be daunting to adopt. In this post I’ll go through “crawl, walk, run” steps for making

中文介绍 介绍软件工厂模型采用步骤：爬行、步行、奔跑。

How we built Hermes to support our entire team

@JacquelineSYC19 · 1.0K 粉丝 · 72.4K 阅 · 501 赞 · 47 转

09/17 00:35

At the time of writing, Artie is a team of 17. Every one of us works alongside Hermes, an open-source AI agent harness from @NousResearch. It runs on a single physical machine in Germany, has a

中文介绍 分享构建支持团队工作的开源AI代理Hermes的经验。

The Anatomy of Harness Engineering: How to Evaluate, Iterate, and Guard AI Coding Agents

@GoogleCloudTech · 1.3M 粉丝 · 43.8K 阅 · 505 赞 · 60 转

09/16 03:41

When developers first work on harness engineering for agentic coding systems, they often fall into the same trap: they run common end-to-end benchmarks like Terminal-Bench and DeepSWE, watch a

中文介绍 解析AI编码代理工程：评估、迭代和防护。

Introducing the DeepMind Institute

@ShaneLegg · 82.3K 粉丝 · 40.6K 阅 · 585 赞 · 86 转

09/16 22:25

We are on the cusp of a profound transformation. Today’s AI systems have impressive capabilities and the rapid pace of innovation suggests we’re now approaching artificial general intelligence (AGI),

中文介绍 介绍DeepMind Institute，展望迈向通用人工智能的进程。

Build real-time voice applications with Gemini 3.8 Live and 3.5 Transcribe

@GoogleAIStudio · 203.4K 粉丝 · 35.5K 阅 · 506 赞 · 51 转

09/16 05:09

New Gemini Audio models are available for developers to build more intelligent conversational experiences via the Gemini API and Google AI Studio. Today, we released new Gemini Live models in the

中文介绍 介绍Gemini 3.8 Live和3.5 Transcribe，构建实时语音应用。

Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking

@GoogleAIStudio · 203.4K 粉丝 · 35.1K 阅 · 585 赞 · 47 转

09/16 01:05

Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking are our most advanced live dialogue models yet. Major upgrades in intelligence and parallel reasoning make it more intuitive to collaborate and

中文介绍 介绍Gemini 3.8 Live和3.8 Live Extended Thinking，提升智能对话能力。

AI Agents Are Moving Beyond Chat

@Didotxyz_ · 64.6K 粉丝 · 5.7K 阅 · 502 赞 · 325 转

09/17 03:05

The important shift in AI agents is not the disappearance of chat. It is that chat is no longer the whole product. An assistant explains a task. An agent system has to carry it forward: call a tool,

中文介绍 AI代理发展超越聊天，转向任务执行和工具调用。

[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 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推出Astra for Law，提供法律领域的先进智能和定制化工作流程。

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，利用税收AI。

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首席执行官谈其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分享模型偏差报告框架，包括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合作，为1000名美国老年人提供免费的ChatGPT研讨会，以安全地构建实用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 OpenAI探索新的AI驱动广告体验，包括赞助代理、营销工具和与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，半导体和数据中心的性能接近物理极限。

How to connect AI usage to business value

OpenAI News · 09/16 20:00

Learn how ChatGPT Work and Codex analytics help teams understand AI usage and spend, identify training needs, and connect adoption to business outcomes.

中文介绍 学习如何通过ChatGPT Work和Codex分析连接AI使用与业务价值。

[AINews] Jev: a “System One Model” that only decides/classifies/routes/scores — >100x faster, >200x cheaper than small frontier LLMs

Latent Space · 09/16 19:09

congrats to TypeSafe!

中文介绍 Jev是一个“系统一模型”，决策/分类/路由/评分速度快于小型前沿LLM超过100倍，成本低200倍以上。

How workers are unlocking new ways of working

OpenAI News · 09/16 17:00

New OpenAI Economic Research shows how workers use AI beyond traditional roles and which new activities become recurring parts of their work.

中文介绍 OpenAI经济研究显示，工人如何使用AI超越传统角色，哪些新活动成为他们工作的常规部分。

Jev ⚡, Periodic Neon 🧬, Gemini 3.8 Live 💬

TLDR AI · 09/16 08:00

中文介绍 Jev、周期性霓虹灯、Gemini 3.8 Live。

Can Skills Learned in Games Transfer to Real-World Work?

Latent Space · 09/16 04:11

Good Start Labs trained an AI on a railroad game — and one version improved at financial research. The difference was the training design.

中文介绍 Good Start Labs训练AI在铁路游戏上，一个版本在金融研究上有所改进，差异在于训练设计。

Roundtables: Could AI really kill us all?

MIT Tech Review AI · 09/16 01:47

Listen to the session or watch below Employees at the world’s leading AI labs are saying there’s a real possibility that advanced AI could destroy humanity. Are they right? Or is this more scaremongering and hype? Watch a conversation unpacking AI extinction fears: where they come from, whether they

中文介绍 世界领先AI实验室的员工表示，高级AI有可能毁灭人类。这是真的吗？还是这只是恐慌和炒作？

Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking

DeepMind Blog · 09/16 01:05

中文介绍 介绍Gemini 3.8 Live和3.8 Live Extended Thinking。