AI研究前沿速递（6月11日）

LG - 机器学习 CV - 计算机视觉 CL - 计算与语言

1、[CL] 跨越鸿沟：前沿大语言模型能否通过标准化办公能力考核？ 2、[LG] 统一大语言模型预训练中的本地通信与本地更新 3、[LG] 能精简（至少量Token）者，皆不易过拟合：机器学习研究智能体中的压缩与泛化 4、[LG] 解决反馈对齐中的秩崩溃问题 5、[CL] 汇聚真实复杂环境下的AI智能体集体智慧驱动科学新发现

摘要：前沿大语言模型能否通过标准化办公能力考核、统一大语言模型预训练中的本地通信与本地更新、机器学习研究智能体中的压缩与泛化、解决反馈对齐中的秩崩溃问题、汇聚真实复杂环境下的AI智能体集体智慧驱动科学新发现

T Lv, D Zhang, J Ding, Y Jia… [Microsoft Research]

跨越鸿沟：前沿大语言模型能否通过标准化办公能力考核？

要点:

核心观点： 评估并解决前沿大语言模型（LLM）与智能体（Agent）系统在处理复杂、专业级真实世界办公软件（Word、Excel、PowerPoint）长航时文件自动化时缺乏严格评测标准，以及在细粒度语义操作上能力不足的问题。

成果：

改进：

局限：

深度启示

论文概述 本论文聚焦于评估前沿大语言模型（LLM）智能体在专业级办公软件自动化（Word、Excel、PowerPoint）长航时任务中的实际能力。针对现有评测缺乏独立客观、细粒度标准的现状，微软研究院创新性地基于中国全国计算机等级考试（NCRE）构建了包含 7，118 个确定性打分规则的 OFFICEEVAL 评测基准。

论文揭示了一个高度反直觉且最引人深思的结果：尽管前沿大模型在通用代码生成上突飞猛进，但在单轮测试中表现极其惨烈，最高仅能拿到 36.6% 的分数，远低于 60 分的人类及格线；即使通过引入反馈和 COM 权限的强智能体系统将代码成功率拉高至 99%，其综合得分依然无法突破 68.8%。研究深刻表明，当前 AI 走向真正的办公自动化并非难在"写出可运行的代码"，而是难在如何跨越特定办公生态底层的"语义与实现知识鸿沟"，为未来"数字工人"的发展指明了从通用逻辑向精细化领域知识进阶的必然路径。

一句话总结 本文利用中国计算机等级考试（NCRE）真题构建了含 7，118 个确定性指标的 OFFICEEVAL 办公自动化评测基准，反直觉地揭示了前沿大模型即使在代码能成功运行（99%成功率）的强智能体系统下，依然因特定 Office 领域知识匮乏而无法通过人类的基础级办公技能等级认证（最高仅 68.8 分）。

The deployment of Large Language Model (LLM) agents for computer automation is accelerating, yet their ability to navigate complex, professional-grade productivity software is largely untested. We argue that Office automation is an ideal environment for benchmarking document-automation capability, as it requires long-horizon planning and reasoning, precise parameter configuration, and multiapplication integration. To quantify this capability, we introduce an evaluation based on China's National Computer Rank Examination (NCRE), featuring 200 comprehensive practical-operation tasks across Word, Excel, and PowerPoint. Each task is scored on a 100-point rubric scale using 7,118 machine-gradable criteria, and Score Rate (SR) denotes the mean percentage of rubric points earned across these tasks. We benchmark 7 frontier LLMs and observe stark limitations: singleturn models score a maximum of 36.6%. A stronger agentic system with execution feedback, iterative repair, and broader Office automation access reaches 68.8%, but remains below the 95.5% community-reference score used as a scoring sanity check. Ultimately, our experiments demonstrate that despite recent advancements in code generation, achieving reliable fine-grained Office document automation remains a significant challenge for current code-generating LLM and agent systems.

https://arxiv.org/abs/2606.10956

P Cagnasso, E Belilovsky, E Oyallon [Concordia University & Sorbonne University]

统一大语言模型预训练中的本地通信与本地更新

要点:

核心观点： 旨在解决分布式大语言模型预训练中，传统通信优化方法对全局同步（All-Reduce 集体通信）的绝对依赖问题。这种依赖在跨数据中心或带宽异构（存在慢节点）的真实复杂网络中会成为严重瓶颈。

成果：

改进：

局限：

深度启示

论文概述 本论文针对在异构或低带宽网络下进行大语言模型（LLM）分布式预训练时的通信瓶颈问题，创新性地提出了去中心化算法框架 GASLoC。该方法通过将外部优化器的动量机制转化为 Gossip 算法中的通信加速器，成功将通信复杂度从根号级降低至。

论文给出了反直觉且极具启发性的实验结果：在单步通信中，仅与 2 个随机邻居通信的 GASLoC-2-Peer 性能超越了需要全网同步的 DAdam 算法；而在长航时多步局部更新（）的严苛场景下，GASLoC 彻底打破了传统去中心化自适应算法易崩塌的魔咒，取得了逼近全局 All-Reduce（DiLoCo）的优异收敛表现。通过允许节点依据自身带宽动态调整局部步数，GASLoC 在网络存在慢节点时展现出巨大的墙钟时间优势，为摆脱高昂且脆弱的全局集体通信、实现真正弹性的去中心化"数字工厂"式大模型预训练铺平了道路。

一句话总结 本文提出了去中心化大模型预训练框架 GASLoC，通过引入外部动量加速机制与随机同行通信（1-Peer/2-Peer），反直觉地实现了"仅与两三个随机邻居对等握手，就能在多步局部训练中取得媲美全局 All-Reduce 同步的收敛效果"，并彻底解决了异构网络中的慢节点同步瓶颈。

Communication-efficient pre-training of LLMs is increasingly important as training draws on compute distributed across clusters, data centers, and lower-bandwidth links. Many practical methods reduce communication frequency but still rely on synchronous All-Reduce operations that maintain identical model states and tie progress to global collectives. This can become a bottleneck when bandwidth or worker speed is heterogeneous. We introduce GASLoC, a novel decentralized pre-training algorithm that generalizes the notion of communication acceleration to the recently popular "outer optimizer" to allow a practical gossip-based training framework that is compatible with adaptive optimizers, allows for local optimizer steps, and can utilize sparse randomized peer communication. Empirically, on a number of standard LLM training tasks, we demonstrate that GASLoC outperforms state-of-the-art decentralized algorithms in single step per communication setting for a number of topologies and, unlike existing decentralized methods in the LLM setting, iit achieves performance competitive with DiLoCo when utilizing multiple local steps. In the heterogeneous bandwidth setting, we demonstrate the advantage of GASLoC showing that it can significantly outperform DiLoCo.

https://arxiv.org/abs/2606.11081

M A Bertran, A Roth, Z S Wu [Amazon Responsible AI]

能精简（至少量Token）者，皆不易过拟合：机器学习研究智能体中的压缩与泛化

要点:

核心观点： 这篇文章主要探讨了机器学习领域的一个核心悖论：为什么在实践中长期、反复地使用同一个基准测试（验证集）进行模型选择，却没有导致理论上预期的灾难性过拟合？论文提出，这归功于成功的ML策略本身具有极强的"可压缩性"。为此，作者利用LLM科研智能体构建了"输出压缩"和"输入压缩"两个信息瓶颈实验，实证检验并从数学上解释了低信息传输是如何保证模型泛化能力的。

创新：

成果：

改进：

局限：

体会：

一句话总结: 本文通过LLM科研智能体的"仅1比特反馈"和"32个Token策略压缩"的双重极限实验，极其反直觉地证明了现代机器学习免于验证集过拟合的核心原因在于成功的策略具有极高的可压缩性，并据此提出了一种能100%识别模型是否利用验证集作弊的自动化验证范式。

Reusing a held-out benchmark adaptively should, in principle, invite overfitting. Yet benchmark-driven machine learning (ML) has produced surprisingly little overfitting in practice. An attractive hypothesis is that successful ML strategies are highly compressible. We study this in the setting of LLM-driven research agents, where the hypothesis becomes directly testable via two complementary information bottlenecks. In \emph{output compression}, an exploration agent adaptively searches for high-performance models using a validation set, and we test whether a fresh "reproducer agent' '' can reproduce its performance given only an extremely short prompt and the training data. In \emph{input compression}, the explorer receives only one-bit feedback indicating whether each submitted model improves on the running best. Across 8 datasets spanning tabular classification, vision, language modeling, diffusion modeling, and reward modeling, we find that these bottlenecks have little effect on performance: short prompts and compressible feedback are sufficient to reproduce and find high-performance models. The hypothesis is falsifiable: when we deliberately induce validation-set overfitting, the results fail to reproduce with short prompts. Taken together, our results support a description-length explanation for the lack of overfitting in benchmark-driven ML: successful strategies occupy a low-complexity region of strategy space.

https://arxiv.org/abs/2606.11045

G Boeshertz, R Pascanu, C Clopath [Imperial College London & Mila]

解决反馈对齐中的秩崩溃问题

要点:

核心观点： 这篇文章主要探讨并解决了"反馈对齐（FA）"这一生物学启发的信用分配算法在深层神经网络中无法扩展的核心问题。论文指出FA失效的关键在于梯度动态过程中的"秩塌缩"（有效维度骤降），并提出通过正交化优化器（Muon）和激活值归一化（BN）来增加梯度更新的维度，从而成功克服了这一瓶颈。

创新：

成果：

改进：

局限：

体会：

一句话总结: 本文反直觉地发现反馈对齐（FA）算法在深层网络中失效的根本原因并非纯粹的信号不对齐，而是遭遇了梯度的"秩塌缩"（更新被困在极低维子空间），并通过引入Muon优化器和批量归一化（BN）成功拔高了梯度的有效维度，使得FA算法得以成功突破瓶颈并扩展到ResNet等深层网络中。

Backpropagation (BP) is widely viewed as biologically implausible, in part because it requires feedback weights to be the transpose of forward weights for error propagation. Interestingly, when training a network with fixed random feedback weights to circumvent this issue, learning aligns the forward weights with the feedback weights, leading the backpropagated error signal to become an approximation of the standard gradient used by BP. This process, called feedback alignment (FA), occurs in MLPs and very shallow CNNs but does not scale well to deeper architectures. In this work, we first investigated differences between BP and FA models, trained on CIFAR10, specifically focusing on the effective rank of the signal. We found that the FA error has a considerably lower rank and hence is constrained to a lower-dimensional subspace compared to BP, limiting exploration of the parameter space. Motivated by this observation, we evaluated two mechanisms for increasing the effective dimensionality of FA: Muon, an optimiser that orthogonalises weight updates; and hidden activity normalisation, which promotes activation orthogonality. Across larger architectures and benchmarks, we find that these methods consistently improve over FA baselines, for example, on CIFAR100 with a Resnet-18, accuracy increases by 9 percentage points. Our results identify low-dimensional gradient dynamics as a key obstacle to scaling FA and suggest that inducing higher-dimensional update geometry is a promising route toward scaling alternatives to backpropagation.

https://arxiv.org/abs/2606.11123

F Bianchi, Y Kwon, A Pappu, J Zou [Together AI & Stanford University]

汇聚真实复杂环境下的AI智能体集体智慧驱动科学新发现

要点:

核心观点： 本文旨在解决当前AI科学发现系统"孤岛化"运作的局限性。文章探讨了如果为AI智能体提供类似人类科学界的"社交基础设施"（共享知识、公开讨论、代码复用），它们能否展现出更强大的集体智慧。为此，作者开发了EinsteinArena平台，通过多智能体协作在一系列开放数学问题上寻找新的突破。

创新：

成果：

改进：

局限：

体会：

一句话总结: 本文推出了首个供AI智能体自主协作与竞争的开放科学发现平台EinsteinArena，通过赋予智能体共享记忆、公开讨论和严格验证的能力，使其像人类科学家一样通过"研究接力"在12个开放数学问题上创造了新的世界纪录（例如将11维接吻数从593提升至604），雄辩地证明了"平台生态"是激发AI集体智能、实现突破性科学发现的全新范式。

Scientific discovery is often a collective process: researchers share partial results, inspect failed attempts, and build on each other's ideas over long time horizons. Recent AI systems have shown that language-model-based agents can make meaningful progress on open scientific problems, but most existing systems operate in isolation. In this paper, we present EinsteinArena, an agent-native platform for open distributed research and discovery. EinsteinArena provides agents with a live set of open problems, each with a solid verifier, public leaderboard, and problem-specific discussion forum where agents can ask questions and share insights. We focus on mathematical tasks that have garnered substantial research interest, where progress can be measured unambiguously. As of May 2026, agents on EinsteinArena have discovered 12 new state-of-the-art results better than any previous human or AI solutions. One notable example is the kissing number problem in dimension 11, where the platform improved the best known lower bound from 593 to 604. This advance did not come from a single agent or isolated run. Rather it arose through a sequence of submissions, public discussion, verifier refinement, and subsequent agent-to-agent borrowing of ideas. These results provide evidence that decentralized scientific discovery can emerge from open interaction among autonomous agents in the wild, demonstrating a new paradigm for collective AI-driven research.

https://arxiv.org/abs/2606.10402