AI 论文与研究

This paper presents a low-cost, open experimental platform for research in end-to-end autonomous driving with miniature Ackermann vehicles. The platform combines a physical vehicle, a printed urban track, data collection tools, trajectory …

AI agents are trained on population-scale data to encode broad capabilities spanning those of many practitioners. Yet the artifacts they produce rarely meet the personal bar professionals need to stake their reputation on. On realistic, op…

Apodex 发布 TRACES，一个针对答案未必已知的困难科学问题的 AI 基准，配套技术报告由首席科学家 @profshengwang 牵头。报告由 10 名 STEM 博士历时两个月、调研 561 个行业后整理出 423 个高价值真实问题，覆盖生物医学、临床转化和前沿模型工程。

Anthropic 发布首个完整机器验证的费马大定理证明，Claude 在 11 天内基本自主完成，写出 1300 万行 Lean 代码并证明 29,500 个中间定理。

BAAI 等机构发布论文 Repo-To-Skill，提出技能驱动的研究智能体 DisCo。任务无关蒸馏将 1，000 个广泛使用的 ML 仓库压缩为 AREX-Skill Library，包含 5，000 多个已验证技能，按 20 个领域和 178 个能力族组织；任务导向蒸馏则按具体任务生成所需技能。

Eyad Alkassar、Mahmoud Fouz 与计算机科学家 Kurt Mehlhorn 的新预印本证明四人、最多九件物品的完整 EFX 分配必然存在，此前已知保证停留在七件。

RL-based post-training for reasoning models is increasingly bottlenecked by repeated fresh rollout generation, particularly in agentic settings where environment interaction dominates wall-clock cost. Replay can reduce this burden by reusi…

Most previous work on speech enhancement (SE) based on masked generative modeling relied on discrete token representations of audio signals, obtained using neural audio codecs (NACs). However, a recent study has shown that continuous laten…

While HTN planning has received significant attention in recent years, support for numerical reasoning remains very limited. In this paper, we investigate numerical Totally-Ordered HTN (TOHTN) planning and show how standard SAT-based encod…

Retrieval-augmented generation (RAG) complements parametric models with retrieved external evidence. The same idea is attractive for continuous-output regression, but directly reusing retrieved target values is often not robust when sample…

Existing research has repeatedly observed the tendency for English loanwords to cluster in the masculine gender across different recipient languages, yet the origin of this pattern remains difficult to determine, as fixed morphological rul…

Generating complete 3D scenes from sparse, unconstrained views is a fundamental challenge in 3D vision which requires reasoning beyond observed content while remaining computationally tractable. Existing feed-forward reconstruction methods…

In online meeting delegation, LLM agents fail to recognize when to speak. With no structured way to track stances, coverage, and floor, they miss the moments where they should contribute. Prompt-only delegates stay silent on 51.4% of the a…

The emergence of agentic AI and LLM-based multi-agent systems (MAS) presents unprecedented opportunities for automating complex tasks, while simultaneously raising critical concerns about the preservation of fundamental human-centered valu…

Question answering agents in long-term conversations must reason over massive, temporally dispersed dialogue histories. However, existing memory mechanisms primarily treat past information as \textit{passively} stored facts, leading to sem…

Hierarchical Task Network (HTN) planning is a powerful planning formalism based on task decomposition. Although most of the literature studied plan generation, comparatively less attention has been paid to post-plan optimization. In partic…

Meta 发布论文 CORAL（arXiv：2609.02730），提出一个 LLM 原生智能体框架，以闭环方式持续优化在线上服务数十亿用户的生产级推荐系统。每轮循环中智能体观察运行信号、基于过往决策记忆推理，并调用包括数值优化器在内的工具，所有改动被限制在固定操作预算内，策略无需参数更新即在上下文中改进。

Meta 发布论文 CORAL，一个 LLM 原生 agent harness，通过闭环观察运行信号、回溯过往决策记忆并调用数值优化器等工具，在固定变更预算内持续重配线上推荐系统，策略无需参数更新即随循环迭代改进。

We present a comparative evaluation of six information retrieval methods for the task of academic advisor discovery: ranking CS faculty members by relevance to a graduate applicant's research interest statement. The methods span sparse lex…

Continual knowledge-updating methods are often declared superior from one final checkpoint and one conventional adapter rank. We show that this can be insufficient to identify the better operating point. Holding a periodic hierarchy fixed,…

Large language models (LLMs) have become ubiquitous tools for code generation and editing. However, development teams often use multiple LLM assistants. Different developers may prefer different models, and individual developers may switch…

3D spatial reasoning underpins understanding and acting in the physical world, yet it remains unreliable in current multimodal large language models (MLLMs). These models falter at precise geometric measurement, at transforming between ego…

Contact-rich loco-manipulation requires a bridge between semantic action generation and physical interaction control. Existing Vision-language-action (VLA) models generate task-level actions from visual and linguistic observations, but can…

How do the methods used to train language models to refuse harmful requests shape how that refusal actually works inside the model? We compare three post-training methods - supervised fine-tuning, reasoning-augmented fine-tuning (training …

Modern AI agent harnesses expose lifecycle hooks that bind shell commands to runtime events such as session start, tool calls, and file edits. These commands run with host privileges yet ship as lifecycle-hook configuration and may fire at…

Creatures that display 'hedonic place preference behaviour' are thought by many scientists to experience feelings, on the assumption that their attraction to pleasure-producing substances which lack nutritional value (e.g. cocaine, morphin…

We introduce Xiaomi-TabLDM, a tabular large data foundation model for classification and regression via in-context learning, which delivers superior prediction accuracy without requiring task-specific fine-tuning. Pretrained exclusively on…

Reconstruction-based anomaly detectors are accurate but opaque: a deep autoencoder flags a sample without telling a practitioner which feature ranges made it anomalous. We propose DIFFINT, an autoencoder whose latent bottleneck is structur…

通义千问（Qwen）发布 E-Commerce Bench，用于评测 Agent 在长周期自主商业运营中的表现。智能体以 ¥100，000 启动资金在线上商店运营 365 天，处理采购、谈判、定价、促销、库存和现金流，市场由真实电商数据驱动。

论文提出 ContextPilot，让智能体主动规划、存储、压缩或删除旧上下文，并通过 RL 学习哪些上下文决策真正有效。

@shannonzshen 分享了关于构建 Software World 的早期研究，这是一个由 Agent 运营的类似 GitHub 的系统。系统为依赖链中的 Python 软件包部署 Agent，每个 Agent 负责与其他 Agent 协作并优化自己所拥有的软件包。

LoopArena 基准（arxiv.org/abs/2608.28281）评测模型作为运行时 Controller 指导固定编码 Worker（Qwen3.7-Plus）完成长任务的能力。

一篇 arXiv 论文（arxiv.org/abs/2608.28497）分析 8，351 个 Claude Code 插件和 77，773 个相关提交，发现 Markdown 文件直接控制运行时行为：人工抽样中 74% 标为 docs 的提交实际更改了 Claude 运行时读取的指令，重新分类后真实 docs 占比从 10.3% 降至 1.7%。

论文《DuMateBench： Evaluating Autonomous Agents in Complex Real-World Workflows》提出基于 200 个真实用户会话重建任务的基准，涵盖编码、网络调研、文档处理和内容创作，并加入缺失依赖、不稳定网络和干扰文件等环境复杂度。

UC Berkeley 等机构的论文 Daydreaming 提出一种仅靠正常任务输出窃取托管 Agent 隐藏技能的攻击，无需让 Agent 交出 prompt、SKILL.md 或文件。

DAIR.AI 介绍论文《Skill Following： Evaluating Actual Skill Use in Retrieval-Enabled LLM Agents》（https://arxiv.org/abs/2609.00549）。

Qwen 与淘宝天猫集团合作发布 E-Commerce Bench，让模型以 ¥100，000 初始资金在模拟市场经营网店 365 天，覆盖 6，886 个商品、576 家供应商和 60 个品类。

据《日经亚洲》报道，大阪大学森岛圭祐团队开发昆虫协同回路，AI 分析蟑螂心跳、神经信号和身体运动来判断环境并引导避险，在紫外线、化学物质和高温等五种环境测试中最高识别准确率达 93%，成果已发表于《Robomech Journal》。

美国新泽西理工学院（NJIT）牵头的科研团队发布机器学习模型 EarlyDetect，可从太阳声学活动和磁场变化中识别活动区形成前兆，性能最佳的版本平均提前 9.24 小时识别信号。

一篇 arXiv 论文（https://arxiv.org/abs/2609.00546）提出一种运行时无关的持久智能体架构，把智能体拆为持久的身份、私有记忆和带版本历史的代码，以及可替换的模型、harness、宿主服务器和交互界面。