跳到主内容
@wquguru
精选88DAIR.AI(RSS)论文研究

DAIR.AI周报:Meta CLM、MIT JAZ与微软Agensh等5篇A

🥇Top AI Papers of the Week

原文
发到 X

1. Context Language Models

1. 上下文语言模型

Agent harnesses usually manage the model’s context through fixed rules such as summarization or compaction. Researchers at Meta and collaborators propose Context Language Models (CLMs), which treat the live context as a file the model can edit freely, deciding what to keep, rewrite, or remove.

智能体框架通常通过摘要或压缩等固定规则来管理模型的上下文。Meta 及其合作者提出上下文语言模型(CLMs),将实时上下文视为模型可以自由编辑的文件,由模型自行决定保留、重写或删除哪些内容。

  • Context as a read-write file: The model edits its own context with code, writing loops that prune irrelevant results, defining functions it reuses, and tracking subagents. Recursive language models place a large input in an external variable the model can read, but they do not let it edit its live interaction context. Because several agent contexts can coexist as files, the approach extends to multi-agent systems.
  • Works zero-shot: Built from existing models with no training, CLMs reach 11.4% higher accuracy with 21.5% fewer FLOPs on BrowseComp-Plus than state-of-the-art context-management strategies, 5% higher scores with 59% fewer FLOPs on the 12-hour EdgeBench, and 65% more improvement for the same compute on a 24-hour agent-swarm task across six repositories.
  • Learnable in context and in weights: Natural-language instructions evolved through a skill-optimization loop raise held-out accuracy by up to 35.9 points on a context-management task. An online RL method improves Qwen3.5-9B on BrowseComp-Plus by 47.6% while using 12% fewer FLOPs.
  • Why it matters: Edits in the middle of the context break prefix caching, so the authors co-design Suffix Cache Reuse, which cuts server-side compute by 35% against standard SGLang at matched performance. Context management can move from the harness into the model without raising serving cost.
  • 上下文作为读写文件:模型通过代码编辑自身的上下文,例如编写循环以剪枝不相关的结果、定义可复用的函数以及跟踪子智能体。递归语言模型将大型输入置于外部变量中供模型读取,但不会允许其编辑实时交互上下文。由于多个智能体上下文可以共存为文件,该方法可扩展至多智能体系统。
  • 零样本工作:CLM 基于现有模型构建且无需训练,在 BrowseComp-Plus 上相比最先进的上下文管理策略准确率提高了 11.4%,浮点运算次数(FLOPs)减少了 21.5%;在 12 小时的 EdgeBench 测试中得分提高 5%,浮点运算次数减少 59%;在跨越六个仓库的 24 小时智能体集群任务中,相同计算量下性能提升了 65%。
  • 可在上下文中和权重中学习:通过技能优化循环进化的自然语言指令,使上下文管理任务的未见过数据准确率最高提升 35.9 分。一种在线强化学习方法使 Qwen3.5-9B 在 BrowseComp-Plus 上的表现提升了 47.6%,同时浮点运算次数减少了 12%。
  • 重要意义:对上下文中间部分的修改会破坏前缀缓存,因此作者联合设计了后缀缓存复用(Suffix Cache Reuse),在保持与标准 SGLang 相当性能的同时,将服务器端计算量降低了 35%。上下文管理可以从框架移至模型内部,而不会增加服务成本。

Paper | Tweet

论文 | 推文

2. JAZ

Long-term memory and self-improvement are usually built as separate systems around the agent loop. Researchers from MIT CSAIL built JAZ to test how far a minimal harness, little more than the agent loop itself, can go on the tasks those systems target.

长期记忆和自我改进通常作为围绕智能体循环的独立系统构建。麻省理工学院计算机科学与人工智能实验室(MIT CSAIL)的研究人员构建了 JAZ,旨在测试一个极简框架——仅比智能体循环本身多一点——能在这些系统所针对的任务中走多远。

  • One primitive: JAZ exposes a single LLM-based primitive, invoke, which acts as a function whose implementation the LLM writes at runtime each time it is called. Built-in hooks let the programmer apply constraints and monitor the run.
  • Everything is a variable: The LLM can write arbitrary executable code, including recursive calls to invoke, so subagents are the default. All inputs to invoke and the full interaction history live as variables in the code environment, where the model can read and transform them with code.
  • Beats specialized systems: With prompting only and no manually designed tools, memory system, or file system, JAZ outperforms Letta (MemGPT) by 8% at half its cost on the recall-heavy portion of StuLife, a long-horizon benchmark that requires recall far beyond the context window. On AppWorld, it beats ACE, a continual self-improvement method, by 4% at lower cost.
  • Why it matters: Memory and self-improvement become code the agent writes inside its own loop. If you build custom harnesses, this is evidence that a smaller surface can match heavier infrastructure on these workloads.
  • 单一原语:JAZ 暴露了一个基于大语言模型的单一原语 invoke,它充当一个函数,其实现由大语言模型在每次被调用时在运行时编写。内置钩子允许程序员施加约束并监控运行过程。
  • 一切皆变量:大语言模型可以编写任意可执行代码,包括对 invoke 的递归调用,因此子智能体成为默认设置。invoke 的所有输入及完整的交互历史均以变量形式存在于代码环境中,模型可以通过代码读取和转换它们。
  • Beat 专用系统:仅通过提示且无需手动设计的工具、记忆系统或文件系统,JAZ 在 StuLife(一个需要远超上下文窗口范围的回忆能力的长周期基准测试)的回忆密集型部分以一半的成本超越了 Letta (MemGPT) 8%。在 AppWorld 上,它以更低的成本超越了 ACE(一种持续自我改进方法)4%。
  • 为何重要:记忆和自我改进成为智能体在其自身循环中编写的代码。如果你构建自定义框架,这表明在这些工作负载下,较小的表面可以匹敌更重的基础设施。

Paper | Tweet

论文 | 推文

3. Agensh

Multi-agent harnesses usually depend on a central orchestrator that assigns tasks and coordinates workers, and that orchestrator limits how many agents the system can use. Microsoft Research introduces Agensh, a self-organized multi-agent harness with no central orchestrator, and scales it to 1,024 coding agents.

多智能体框架通常依赖于一个中央协调器来分配任务和协调工人,而该协调器限制了系统可以使用的智能体数量。微软研究院推出了 Agensh,这是一个无中央协调器的自组织多智能体框架,并将其扩展至 1,024 个编码智能体。

  • A shared cooperation loop: Every worker runs the same loop asynchronously. It gathers context, claims and self-assigns a sub-task, takes action and shares findings, verifies the result, and merges progress.
  • Organization infrastructure: Three components support the loop. A shared workspace holds proposed, ongoing, and completed work, a message interface lets workers communicate, and a shared context retains reusable findings and work intentions.
  • Scaling the number of agents: On the five hardest ProgramBench tasks with GPT-5.6-sol, going from 1 to 128 agents raises the mean final test-pass rate from 19.31% to 28.78%, about a 49% relative improvement, and larger teams reach a given pass rate sooner. On building pandoc from scratch under a 6-hour budget, going from 1 to 1,024 agents raises the test-pass rate from 33.89% to 55.06%.
  • Why it matters: Worker trajectories show forms of cooperation that the agents start on their own and that become standard practice as the organization grows. The authors treat the number of agents as a new scaling dimension, which is most useful for complex tasks under hard latency limits or time budgets.
  • 共享协作循环:每个工人都异步运行相同的循环。它收集上下文,认领并自行分配子任务,采取行动并分享发现,验证结果,并合并进度。
  • 组织架构基础设施:三个组件支持该循环。共享工作区保存提议的、进行中和已完成的工作,消息接口允许工人之间通信,共享上下文保留可重用的发现和意图。
  • 扩展智能体数量:在使用 GPT-5.6-sol 的 ProgramBench 五个最难任务上,智能体数量从 1 增加到 128,平均最终测试通过率从 19.31% 提高到 28.78%,相对提升约 49%,且更大的团队能更早达到给定的通过率。在 6 小时预算内从零构建 pandoc 时,智能体数量从 1 增加到 1,024,测试通过率从 33.89% 提高到 55.06%。
  • 为何重要:工人轨迹显示出智能体自发开始的合作形式,并随着组织规模的扩大成为标准做法。作者将智能体数量视为一个新的扩展维度,这在具有严格延迟限制或时间预算的复杂任务中最有用。

Paper | Tweet

论文 | 推文

4. AutoGym

Training agents with RL requires a gym, meaning a task, an executable environment to attempt it in, and a verifier that reliably separates success from failure. These gyms are still built by hand, saturate as models improve, and get exposed to contamination. Researchers from Amazon AGI present AutoGym, which generates complete gyms from a minimal domain seed or from prior model trajectories.

使用强化学习训练智能体需要一个“健身房”(gym),即一个任务、一个可执行的尝试环境以及一个可靠区分成功与失败的验证器。这些健身房目前仍由人工构建,随着模型的提升而趋于饱和,并容易受到数据污染的影响。亚马逊 AGI 的研究人员提出了 AutoGym,它可以从最小的领域种子或先前的模型轨迹中生成完整的健身房。

  • Blueprint first: AutoGym specifies the valid solution space, environment requirements, and verification criteria before it builds the environment. Solvability is settled during construction, so no unreliable LLM judge has to decide correctness afterward.
  • Difficulty you can steer: Explicit generation parameters control task topology, interaction depth, capability axes, question obfuscation, and distractor composition. Single-pass synthetic tasks tend to be hard only in their phrasing, and these parameters target actual difficulty.
  • Active curriculum: Performance-informed calibration shifts the distribution over those parameters as model capabilities change, and failure analysis feeds recurring capability gaps back into the generation parameters.
  • Why it matters: Across productivity and temporal-reasoning settings, AutoGym produces gyms spanning the capability spectrum, including instances that challenge frontier models. Building RL environments is becoming a core AI engineering skill, and this gives a recipe for gyms that keep pace with the models trained on them.
  • 蓝图先行:AutoGym 在构建环境之前,先明确有效解空间、环境需求和验证标准。可解性在构建阶段就已确定,因此无需依赖不可靠的 LLM 裁判来事后判定正确性。
  • 可控的难度:通过显式的生成参数控制任务拓扑、交互深度、能力维度、问题混淆和干扰项构成。单次生成的合成任务往往仅在表述上显得困难,而这些参数针对的是实际难度。
  • 主动课程学习:基于性能的校准会根据模型能力的变化调整这些参数的分布,而失败分析会将反复出现的能力缺口反馈回生成参数中。
  • 为何重要:在生产力和时序推理场景下,AutoGym 生成了覆盖整个能力谱系的 Gym 环境,其中包括挑战前沿模型的实例。构建 RL 环境正成为一项核心的 AI 工程技能,本文提供了一套能跟上在其上训练的模型步伐的 Gym 构建方案。

Paper | Tweet

论文 | 推文

Message from the Editor

编辑寄语

Our new DAIR Academy lab, Build a Website with a Pi Coding Agent, teaches you to build and refine a responsive one-page website by directing a Pi coding agent. Across 5 labs, you add a practical feature to the same project each time, from the first hero section to responsive styling, an interaction, accessibility, and final polish.

我们的新 DAIR Academy 实验室“使用 Pi Coding Agent 构建网站”教你如何通过指导一个 Pi 编码代理来构建和优化响应式单页网站。在 5 个实验室中,你每次都在同一个项目中添加一个实用功能,从首个英雄区(hero section)到响应式样式、交互、无障碍访问以及最终打磨。

Get Started with Pi

开始使用 Pi

5. CASD

Search-based prompt optimizers such as GEPA propose edits, run fresh rollouts, score them, and keep only the edits that improve a validation metric. Researchers from Microsoft show that this loop may be unnecessary when you already have a corpus of agent trajectories.

基于搜索的提示词优化器如 GEPA 会提出修改建议,运行新的 rollout,对其进行评分,并仅保留那些提升验证指标的修改。微软的研究人员表明,当你已经拥有智能体轨迹语料库时,这种循环可能是不必要的。

  • One pass over the logs: Coding-Agent Skill Distillation (CASD) gives an off-the-shelf coding agent a static corpus of trajectories. The agent writes and runs analysis code to compute corpus-wide statistics, find systematic failure modes, inspect representative episodes, and distill the findings into behavioral rules in one prompt. It needs no environment access and no validation data.
  • Reflection scope: Search-based optimizers reflect on small batches of trajectories at each step, so failures that only show up across the whole corpus stay invisible to them. CASD reads the whole corpus, so it can find those failures.
  • Better and cheaper: Across ALFWorld, tau2-bench retail and telecom, and SpreadsheetBench-Verified, a single CASD pass improves the unoptimized baseline by 16.6 points on average, against 10.9 for GEPA and 5.3 for SkillOpt. Each optimized prompt costs about $1.60, over 22x cheaper than validation-gated search.
  • 一次性遍历日志:Coding-Agent Skill Distillation (CASD) 为现成的编码代理提供了一个静态的轨迹语料库。该代理编写并运行分析代码,以计算整个语料库的统计数据,发现系统性故障模式,检查代表性片段,并将发现结果提炼为单个提示词中的行为规则。它不需要环境访问权限,也不需要验证数据。
  • 反思范围:基于搜索的优化器在每个步骤中仅对少量轨迹进行反思,因此那些仅在整个语料库中才显现的故障对它们来说是不可见的。CASD 读取整个语料库,因此能够发现这些故障。
  • 更好且更便宜:在 ALFWorld、tau2-bench retail 和 telecom,以及 SpreadsheetBench-Verified 上,单次 CASD 迭代使未优化的基线平均提升 16.6 分,而 GEPA 为 10.9,SkillOpt 为 5.3。每次优化提示的成本约为 1.60 美元,比基于验证门控的搜索便宜超过 22 倍。

更进一步:量化金融体系

看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力

进入量化体系 →

相似阅读

关联信息,但可能不是同一事件