👍 67
11/10 08:00
External memory enables LLM agents to improve from experience by distilling trajectories of past tasks into concise lessons for new tasks. However, memory can also lead to memory-induced degradation, where an agent fails on tasks it would solve without memory, often because it follows a lesson that
中文介绍 针对 LLM Agent 中外部记忆导致 memory-induced degradation 的问题,提出 Evidence-Linked Memory 方法。通过在记忆条目中关联可验证证据,使 agent 在复用过往经验教训时能判断其适用性,避免盲目遵循失效规则而失败。实验表明该方法可显著减少因记忆误用造成的任务退化。
👍 63
11/10 08:00
Weight-space programming, introduced in recent MIT work by Takehi et al., enables a frontier LLM agent to read a network's weights and write a repair. Their GPT-6 Astra agent corrected the reasoning loops of Qwen3.5-0.8B without gradients, using 158 agent turns, 89.7 million tokens, and 50 forward-p
中文介绍 提出 Tree-Ising,用位级 Ising 编辑修复语言模型。基于 weight-space programming 思路,让前沿 LLM agent 直接读取并写入网络权重,无需梯度即可完成模型修补:GPT-6 Astra 仅用 158 个 agent 轮次、89.7M tokens 和 50 次前向传播即修复 Qwen3.5-0.8B 的推理循环问题。
👍 41
10/29 08:00
Mathematical supervision for language models is typically flattened into isolated problem-solution pairs, discarding the curricular organization and evidential structure of the original sources. We introduce TOME-Math (Textbook-Oriented Mathematical Evidence), a large-scale corpus that converts 2,11
中文介绍 指出数学监督常把教科书内容压平为孤立题解对,丢失课程组织与证据结构。本文构建 TOME-Math(Textbook-Oriented Mathematical Evidence)大规模语料,将 211 万道数学题保留其来源教材的层级与证据信息,用于提升 LLM 的数学推理与可追溯训练效果。
👍 40
11/08 08:00
Generation, imputation, and prediction query different aspects of the same tabular distribution, but are usually evaluated separately. This leaves unclear which capabilities they share, how their training objectives interact, and whether one model can perform all three well. We introduce the One Tab
中文介绍 针对表格数据中生成、插补、预测三类任务常被分开评估的问题,提出 One Table, Many Queries 基准。该基准让同一模型在统一表格分布下同时回答三类查询,用于分析各任务共享能力、训练目标相互作用,以及单一模型能否在三方面同时表现良好。
👍 32
11/10 08:00
Low-bit quantization reduces the inference cost of LLM agents but can compromise task success. Examining failed rollouts, we find that quantized agents overlook task requirements or environmental conditions, producing environment transitions that diverge from successful full-precision references. Co
中文介绍 提出 SteerDistill,用于 LLM Agent 的低比特量化。分析失败 rollout 发现量化 agent 会忽略任务要求与环境条件,其环境转移偏离全精度参考。方法采用闭环教师引导,让全精度教师模型纠正量化模型的偏差,从而在降低推理成本的同时保持任务成功率。
👍 30
11/07 08:00
Weight-activation quantization can reduce the memory and compute cost of neural-network inference, but recent methods often require expensive optimization and architecture-specific propagation of basis transforms across layers. We introduce MACaQ, an MXFP4 W4A4 post-training quantization method that
中文介绍 提出 MACaQ,一种模型无关、快速的权重-激活联合校准量化方法。针对现有方案依赖昂贵优化与逐层基变换传播的缺陷,MACaQ 实现 MXFP4 W4A4 后训练量化,无需昂贵优化即可降低神经网络推理的内存与计算开销,并具备跨架构通用性。
👍 28
10/29 08:00
Large language models (LLMs) draw on parametric knowledge and retrieval-augmented generation (RAG) to solve knowledge-intensive tasks. Yet problem solving in unfamiliar domains calls for tightly coupled knowledge acquisition and reasoning, beyond factual retrieval and aggregation or learning solely
中文介绍 针对陌生领域任务需要知识获取与推理紧密耦合的问题,提出用强化学习激励 LLM 自主学习。该方法超越单纯事实检索聚合或参数内学习,让模型在知识密集型任务中主动获取并整合所需知识,提升跨领域问题求解能力。
👍 27
11/09 08:00
Early successes on agentic robotics point to a potential paradigm shift toward using coding agents to control robots. Existing robotics benchmarks are built around visuomotor policy learning, leaving it open whether coding agents can reliably control diverse robots under physical-world constraints.
中文介绍 提出 Embodied First Exam 评测,检验 coding agent 能否可靠控制多样机器人。现有机器人基准围绕 visuomotor 策略学习,未能覆盖物理世界约束下的编码式控制。该评测考察 agent 在真实物理限制下执行机器人任务的能力,评估以 coding agent 驱动机器人的范式可行性。
👍 27
11/10 08:00
System-1 Decision Models directly make structured probabilistic decisions over query-specified options given a state in a single forward pass, without autoregressively generating reasoning trajectories or textual responses. Recent evaluations cover diverse tasks but often rely on existing benchmarks
中文介绍 提出 MMDM 基准,用于衡量 System-1 Decision Models 的多任务决策能力。这类模型在单次前向传播中对查询指定选项直接做结构化概率决策,无需自回归生成推理轨迹或文本。MMDM 覆盖多样任务,弥补现有评测在多任务决策评估上的不足。
👍 25
11/05 08:00
Language agents increasingly work for users on the open Web, and finance is where trusting them is hardest. The data must come from the Web, the strategy lives in the user's head, and correctness depends on time, so an agent can fail silently: a strategy the user never asked for, unverified data, or
中文介绍 提出 FINCOSMOS,面向金融领域 agent 的开放世界评测范式。由于金融数据须来自 Web、策略存在于用户脑中且正确性依赖时间,agent 易静默失败(未授权策略、未验证数据)。该范式在开放网络环境中评估 agent 的可靠性与正确性。
👍 21
11/07 08:00
How little of a pretrained large language model (LLM) has to change for it to acquire new capabilities? We find that zeroing fewer than 0.05% of its weights, with no other modification, matches full fine-tuning (FFT) at scale across three LLM post-training regimes: reinforcement learning with verifi
中文介绍 研究发现仅将预训练 LLM 少于 0.05% 的权重置零、不做其他修改,即可在规模化下匹配完整微调(FFT)的效果,并覆盖 RLHF、SFT 等三种后训练范式。该结果揭示 LLM 后训练所需参数改动远小于预期,凸显权重稀疏性的重要性。
👍 18
11/07 08:00
Web-facing AI agents repeatedly access webpages, documents, repositories, and tool responses, yet reading interfaces often expose entire objects when only sparse evidence is needed. This over-reading increases token and latency costs and can dilute relevant evidence. We present SparseRead, a reusabl
中文介绍 面向 Web agent 反复访问网页、文档、仓库等对象时过度读取的问题,提出可复用的 SparseRead 方法。读取界面常暴露完整对象,而实际只需稀疏证据,导致 token 与延迟成本上升并稀释相关证据。SparseRead 通过高效稀疏读取外部内容,实现 read less, solve more。
👍 14
11/06 08:00
Large language models (LLMs) increasingly support Web-based services that must balance response quality and inference cost across heterogeneous requests. Conventional routing selects one model per prompt, while fixed ensembles cannot adapt their composition and cost to individual requests. We introd
中文介绍 针对 LLM Web 服务需平衡响应质量与推理成本的问题,提出 prompt-adaptive 成本感知集成路由。传统路由每 prompt 只选单一模型,固定集成无法按请求调整组成与成本。该方法动态调整集成构成,在不同请求间自适应分配模型以兼顾质量与开销。
👍 13
11/09 08:00
AAA games offer a rich testbed for evaluating agent perception, reasoning, and control. However, evaluating whether agents can complete tasks in AAA games is challenging and often human-annotated because task outcomes are often difficult to verify automatically, let alone applying it for Reinforceme
中文介绍 提出 GTAGym,面向 AAA 游戏的真实任务评测与强化学习环境。由于 AAA 游戏任务结果难以自动验证,评测多依赖人工标注。该环境提供可自动验证的任务结果,使 agent 感知、推理与控制能力可被评估,并支持直接用于强化学习训练。
👍 11
11/09 08:00
Agentic workloads are a major source of LLM inference demand. Existing prefill--decode (PD) disaggregated serving systems offer limited flexibility for runtime adaptation of compute allocation and KV cache management, and rely on coarse signals that only partly capture the diverse and dynamic agent
中文介绍 提出 Evolving Infra,一种大规模 agent 自适应 LLM 服务系统。针对现有 prefill-decode 分离式服务在计算资源分配与 KV cache 管理上缺乏运行时灵活性的问题,系统使用更细粒度的动态信号,以适应多样且变化快的 agent 工作负载。
👍 11
11/01 11:58
A scientific agent can give the same verdict while reading and citing different evidence. BioAgent-DF tests this separation using paired reference and candidate values saved at the same scientific computation site. It restores or transplants those values at their original coordinates, then runs the
中文介绍 提出 BioAgent-DF,审计生物医学科研 agent 的决策来源。由于 agent 可能在引用不同证据下给出相同结论,该方法在同一计算位置保存参考值与候选值,在原坐标恢复或移植这些值后运行 agent,检验其判断是否真正依赖被引用的证据。
👍 9
11/07 08:00
Diffusion language models (DLMs) generate text through iterative denoising, where decisions about which tokens to commit or remask strongly affect output quality. Existing certainty-aware strategies rely on instantaneous confidence or entropy, but these local signals can be misleading on difficult r
中文介绍 针对 diffusion 语言模型迭代去噪中 token 提交与 remask 决策影响输出质量的问题,提出频率感知 remasking 的 test-time scaling 方法。现有确定性策略依赖瞬时置信度或熵等局部信号,在困难问题上易被误导,该方法据此改进 token 承诺策略以提升生成质量。
👍 8
11/04 08:00
We introduce Micius-Flash, a 27-billion-parameter model for engineering with native multimodal input, deep reasoning and a 256K-token context window. We design the Micius data pipeline as a knowledge-tree-driven data flywheel comprising three stages: knowledge tree construction and expansion, data s
中文介绍 提出 Micius-Flash,一个 270 亿参数、面向工程领域的 agentic 模型,支持原生多模态输入、深度推理与 256K token 上下文。其 Micius 数据管线采用知识树驱动的数据飞轮,包含知识树构建扩展、数据筛选与合成三个阶段,形成持续优化的工程知识积累。
👍 6
10/29 08:00
While frontier coding agents can generate high-quality 3D scenes and articulated assets, they struggle to create coherent long-horizon animations. When equipped with general-purpose harnesses, they frequently produce physically implausible outputs, generating objects that interpenetrate, lack physic
中文介绍 提出 DyCraft,一个面向长时程 4D 场景生成的 agentic 系统。虽然前沿 coding agent 能生成高质量 3D 场景与可动资产,但在配备通用 harness 后常产生物体互穿、缺乏物理一致性的不合理动画。DyCraft 旨在生成连贯且物理合理的长时程动画。
👍 5
11/10 08:00
Foundation models such as GPT and Claude have shown strong capabilities in spatial understanding and embodied intelligence. These advances make agentic robotics a promising path toward artificial general intelligence in robotics. However, we find that using foundation models to control robots poses
中文介绍 提出 LEASH,关注 agentic robotics 的安全问题。研究发现用 GPT、Claude 等基础模型控制机器人虽展现了空间理解与具身智能优势,但也带来安全风险。LEASH 旨在通过约束机制让基于基础模型的机器人控制系统在保持能力的同时维持安全运行。