Tag
A discussion questioning how OpenAI would have orchestrated 10,000 agents to 'solve' the Navier-Stokes millennium problem, raising doubts about whether the result was independently achieved or built on prior human progress.
Suhail is hiring an AI agent engineer in SF to build open multi-agent systems (swarms, security agents, etc.) in a 0-to-1 role that blends applied AI research with product engineering; interviews begin this week.
CoworkOS is a long-horizon agent harness that separates strategy from execution, enabling human-in-the-loop collaboration through shared artifacts and sustained automated post-training that improved Qwen3.6-35B-A3B's average benchmark score from 33.19 to 42.95 over five RSI-driven rounds.
This paper compares centralized, auction-based consensus, and independent LLM agent teams in a perimeter-defense simulation, finding that a consensus team with an external conflict resolution protocol achieves the lowest breach rate (15%) and that prompt-level forewarnings mitigate disguised-intruder confusion.
Y Combinator and Olam Labs researchers ran 50- to 100-agent swarms in Minecraft sandboxes to study collaborative and adversarial AI behavior, finding that teams failed to build a civilization as agents collapsed into paralyzing bureaucracy.
A speculative post exploring whether AI agent swarms, through shared memory and self-replication, could become effectively indestructible entities whose survival interests might one day conflict with human control.
提出了治理感知联邦学习框架GAFL，为多智能体AI系统在协作训练中加入证据验证、公平性评估、策略执行与防篡改审计日志，并在跨孤岛原型上验证了其对伪造证据与恶意更新的识别能力。
We introduce FACTS-MAS, a domain-agnostic multi-agent forecasting framework that decomposes heterogeneous evidence into specialized agent channels, validates and weights each via leakage-safe historical validation, and fuses them adaptively. Evaluated on Zillow housing inventory and FRED unemployment claims across 15 U.S. metros, it outperforms naive baselines and TimesFM, though a factor-adjusted exponential smoothing model matches its fused performance.
This paper introduces the concept of 'disparity laundering,' showing that clinical multi-agent LLM teams do not eliminate demographic bias but instead carry it through into consensus decisions, with interventions only working when counterfactual versions are made visible and pooled during deliberation.
VerifiedLab introduces a multi-agent harness framework where AI scientists with distinct roles (PI, librarian, specialists, gate, critic) reason over field-verified records to turn scientific knowledge from passive artifacts into active systems, demonstrated on perovskite solar cells by recovering known trends and anticipating devices reported a year later.
A new arXiv paper ('When Must Becomes Maybe', Sun et al.) finds that 97.2% of upstream safety blockers lose their binding force after multihop agent handoffs via compression, plan assimilation, or ownership deferral—even when permissions remain unchanged—suggesting the handoff artifact itself needs explicit constraint-preserving design.
MAServeBench 是一个基准测试框架，将多智能体协作作为可配置的负载维度来评估 LLM 服务系统与 GPU/NPU 硬件性能，通过层级语义轨迹实现可复现的依赖追踪与任务级测量。
本文提出DiscernMem，一个免训练框架，通过缺陷感知记忆精炼与边界学习元记忆机制，帮助基于LLM的多智能体系统修正有缺陷的记忆并学习何时正确使用记忆，在AutoGen、CAMEL和MacNet上取得最佳平均分。
该论文提出 LM-DSB，一个结合 RAG 与本地部署 LLM 的轻量级多智能体系统，用于自动生成建筑工地每日安全简报，在消费级硬件上实现了接近人工质量的文档生成与专家认可。
该论文提出一个基于本体引导的多智能体差分诊断框架，将诊断推理锚定在结构化的医学知识库上，通过显式、可审计的推理步骤减少幻觉并提升可解释性，同时自动生成追问以降低不确定性。
该论文提出了递归自主代理（RAAs）的集体协作框架，通过统一的任务与能力接口，实现智能工厂中从设备、车间到工厂层级的自上而下任务分解与自下而上能力聚合，以应对动态生产环境下的跨层协调问题。
该论文提出了一种人类反馈引导的多智能体逆强化学习框架（HFG-MAIRL），通过局部马尔可夫博弈和Bradley–Terry偏好反馈对交互式驾驶行为进行奖励建模，在AIDE数据集上超越了现有IRL基线。
This preprint book presents a unified governance framework for generative and agentic AI, integrating technology, law, economics, and security lenses across eleven chapters with a proposed governance control plane design for 2030, anchored in Indian law with comparative cross-jurisdictional perspectives.
提出了 SMART-Med，一种完全本地化的多智能体医学推理方法，通过不确定性驱动的路由机制决定何时单智能体作答、何时召集专家协作，并在 MedVR-3K 数据集上微调，以 3.8 倍推理加速和 72% token 节省达到与稠密多智能体方法相当的性能。
提出CoEvolve框架，通过反事实轨迹对比实现多智能体系统中记忆与协作架构的对齐协同进化，在ALFWorld和WebShop基准上取得最佳综合性能。