Curating the frontier of Artificial Intelligence

Improved techniques for fine-tuning flow models via adjoint matching: a deterministic control pipeline

我们提出了一个确定性伴随匹配框架，该框架为基于流的生成模型制定人类偏好对齐作为最佳控制...

We propose a deterministic adjoint matching framework that formulates human preference alignment for flow-based generative models as an optimal contro...

NeuroAgent: LLM Agents for Multimodal Neuroimaging Analysis and Research

多模态神经影像分析通常涉及复杂的, 模态特定预处理工作流程，需要仔细配置, 质量控制...

Multimodal neuroimaging analysis often involves complex, modality-specific preprocessing workflows that require careful configuration, quality control...

SkillOS: Learning Skill Curation for Self-Evolving Agents

基于 LLM 的代理越来越多地被部署来处理流任务,，但它们通常仍然是一次性的问题解决者，无法从过去的经验中学习……

LLM-based agents are increasingly deployed to handle streaming tasks, yet they often remain one-off problem solvers that fail to learn from past inter...

MASPO: Joint Prompt Optimization for LLM-based Multi-Agent Systems

基于大型语言模型 (LLM) 的多代理系统 (MAS) 在处理复杂的协作任务, 方面表现出了希望，其中代理通常是...

Large language model (LLM)-based Multi-agent systems (MAS) have shown promise in tackling complex collaborative tasks, where agents are typically orch...

Can RL Teach Long-Horizon Reasoning to LLMs? Expressiveness Is Key

强化学习(RL)已被应用于改进大型语言模型(LLM)推理,，但对训练如何随任务扩展的系统研究...

Reinforcement learning (RL) has been applied to improve large language model (LLM) reasoning, yet the systematic study of how training scales with tas...

GlazyBench: A Benchmark for Ceramic Glaze Property Prediction and Image Generation

由于复杂的化学成分，开发陶瓷釉料是一个成本高昂的,、耗时的反复试验过程,，给独立的……带来了沉重的负担。

Developing ceramic glazes is a costly, time-consuming process of trial and error due to complex chemistry, placing a significant burden on independent...

AI co-mathematician: Accelerating mathematicians with agentic AI

我们推出了 AI 联合数学家,，这是数学家可以交互地利用 AI 代理进行开放式研究的工作台。人工智能数学...

We introduce the AI co-mathematician, a workbench for mathematicians to interactively leverage AI agents to pursue open-ended research. The AI co-math...

Uno-Orchestra: Parsimonious Agent Routing via Selective Delegation

大型语言模型 (LLM) 多代理系统通常依赖于严格的编排, 致力于平坦的每个查询路由或手工设计...

Large language model (LLM) multi-agent systems typically rely on rigid orchestration, committing either to flat per-query routing or to hand-engineere...

Position: Embodied AI Requires a Privacy-Utility Trade-off

嵌入式 AI (EAI) 系统正在迅速从模拟过渡到现实家庭和其他敏感环境。然而,最近的EAI所以...

Embodied AI (EAI) systems are rapidly transitioning from simulations into real-world domestic and other sensitive environments. However, recent EAI so...

Executable World Models for ARC-AGI-3 in the Era of Coding Agents

我们评估了 ARC-AGI-3 的初始编码代理系统，其中代理维护一个可执行的 Python 世界模型, 对照之前的 o...进行验证。

We evaluate an initial coding-agent system for ARC-AGI-3 in which the agent maintains an executable Python world model, verifies it against previous o...