AI 论文与研究

一项研究探讨大语言模型在对话生成产物中传播局部修订的能力，推出人工标注基准 RevPropBench，含 120 个测试样本，覆盖九个领域及三种 JSON 元素规模。研究评估九种修订方法，基线准确率为 68.3-93%；从三个并行样本中择优最具性价比，可将准确率提升 2.2-9.7%。

研究探讨 on-policy distillation（OPD）中训练数据的作用，发现仅用 1 条 query 训练即可持续提升数百步并恢复全量数据 OPD 的大部分增益。

2026 PNPL 竞赛聚焦非侵入式语音解码，基于扩展数据集 LibriBrain100（新增 32 名被试，每人约 40 分钟数据，单被试数据增至约 80 小时）推进词汇分类任务。竞赛设两条赛道：Deep track 面向大规模被试内词汇分类，Broad track 则针对跨被试泛化，将被试专属微调数据从约 40 分钟逐步缩减至约 10 分钟，向临床可行的非侵入式脑机接口迈进。

arXiv 论文 FM-Bench（Football Management Benchmark）让 15 个前沿模型在 20 个模拟赛季、约 340 到 400 个决策点的环境中管理足球俱乐部，涉及转会、合同、现金和投资。

DAIR.AI 推荐 arXiv 论文 HarnessEvolve（https://arxiv.org/abs/2609.00829），该工作指出自进化智能体存在三类失败：仅有终态反馈导致错误归因不清、智能体记住任务特定模式而非通用能力、无防护的更新会抹掉已有能力。

Haus Research 对 Perplexity 的 sonar 与 sonar-pro 两个搜索模型提出 310 个关于 210 家科技公司的事实问题，抓取全部 cited URL 核验。

Speech brain-computer interfaces (speech BCIs) translate neural activity into language, offering a path towards restoring speech for people with paralysis and, more broadly, enabling new forms of natural human-computer interaction. Despite…

Recent web agents use world models for test-time action selection by sampling candidate actions, predicting the resulting web states, and ranking them with a ranker model or a Process Reward Model (PRM). These world models are typically tr…

We introduce the Graph Machine (GM), an architecture that maintains an $O(n)$-sized state and accesses it through sparse, dynamic routing. Unlike methods with fixed-size states or sparse but static routing, GM preserves $O(n)$ complexity i…

Ordinary differential equations (ODEs) underlie models in science and engineering, and many applications need derivatives of their solutions with respect to parameters. Ensembles of independent trajectories suit graphics processing units (…

Autonomous robots powered by deep learning face a fundamental auditability challenge: when incidents occur, investigators cannot reconstruct why the system made specific decisions. This paper presents TRACE (Transparent Reasoning Architect…

Harnessing naturally occurring feedback from user interactions offers a promising learning signal for Large Language Models (LLMs). However, recent studies suggest this feedback is inherently noisy and difficult to leverage effectively. We…

We study how far gradient descent (GD) can be accelerated by predetermined stepsizes in smooth convex optimization. Going beyond the classical $Ω(n^{-2})$ first-order oracle lower bound of Nemirovsky and Yudin, we prove an $Ω(n^{-1.6342})$…

LLMs are trained to generate natural language. However, various strands of evidence indicate that an LLM's externalized linguistic outputs and mechanistically-extracted linguistic features can be an unreliable lens for understanding intern…

Competitive programming has become a key test of large language model reasoning, with international competitions such as IOI and ICPC representing its most challenging settings. We present an end-to-end specialization pipeline combining la…

Stable 4-bit floating-point (FP4) pretraining is difficult because the E2M1 payload represents only a narrow range of magnitudes. NVIDIA's Transformer Engine \nv{} recipe addresses this with current-tensor scaling, a randomized Hadamard tr…

Meshfree methods such as smoothed particle hydrodynamics (SPH) with kernel corrections, radial basis function-generated finite differences (RBF-FD), and the local anisotropic basis function method (LABFM) construct discrete differential op…

Researchers increasingly use artificial intelligence to construct measures of social, organizational, and occupational characteristics that are absent from conventional surveys. We propose AICOME, AI COntextual MEasurement, a framework for…

Reinforcement learning with verifiable rewards (RLVR) has emerged as a powerful paradigm for large language model (LLM) post-training, but its reliance on coarse outcome rewards leads to limited guidance on intermediate reasoning processes…

Root cause analysis (RCA) is a critical task in telecom network operations, but diagnosing performance degradations in modern 5G and emerging 6G networks remains challenging due to complex cross-layer dependencies. While large language mod…

For more than 20 years, the Model-RB benchmark frb100-40 remained an open challenge; since 2014, its public record had stood at 99 of 100 variables. We give a directly checkable 100-vertex independent set for its 4,000-vertex graph. Togeth…

People increasingly use language models to support life decisions. Many such decisions involve a probabilistic forecast: How likely is a major life event, a natural disaster, or an economic outcome? Users of language models may implicitly …

Sign language processing systems have traditionally operated at the sentence level, ignoring critical discourse phenomena fundamental to sign language comprehension. We introduce DiscoSign, a computational approach for discourse-aware text…

Generative models have been studied experimentally and theoretically as priors for inverse problems such as compressed sensing. Recent work by Gunn et al. studied the use of generative priors with tunable complexity, where a family of gene…

The performance of LLM-based agents is jointly shaped by the base model and the harness used when interacting with the environment. This exposes them to safety risks in both harmful final responses and multi-step execution trajectories. Ex…

Evaluating LLM agents is essential for guiding their development, yet it has grown prohibitively expensive: a single pass of a frontier model over an agentic benchmark can cost hundreds to thousands of dollars, a price paid repeatedly acro…

Cohere Labs 聚合七个公开目录，构建含 696,291 个工具、123,069 个 MCP 服务器的 Agentic Task Ecosystem 数据集并对外开放。

Google Research 评估了从 UK Biobank 欧洲队列迁移学习到 Biobank Japan 近 20 万日本样本的多基因风险评分（PRS）表现，覆盖 BMI、血压、HDL、LDL、血糖等 8 项临床性状。

Google 与 HHMI Janelia、MRC 分子生物学实验室及剑桥大学等合作，在 Cell 发表雄性果蝇大脑和中枢神经系统的完整连接组，包含超 166,000 个神经元和 1.25 亿个突触连接，是迄今按神经元数量计最大的大脑图谱。

ARC Prize 报告 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 Semi-Private 上，Standard harness 得分 62.7%（成本 $26K），Provider Adapter harness 得分 99.9%（成本 $19K），均为 SOTA。

上海人工智能实验室论文提出 Harness-of-Harness（HoH），把现有 coding agent harness 的执行组织为迭代的规划、编码、测试循环，使智能体在无人工干预下持续改进软件。

Motional 与 MIT 研究人员构建了一套让自动驾驶汽车实时解释自身决策的系统，旨在解决自动驾驶 AI 的黑箱问题。该成果发表于 Nature，团队包括 Motional CEO Laura Major 与 MIT 计算机科学与人工智能实验室的研究人员。

上海人工智能实验室论文提出 Harness-of-Harness（HoH），包裹现有 coding-agent harness，将执行组织为规划、编码、测试的迭代循环，使编码智能体在无人类干预下持续改进软件。

Trellner 于 2026 年 9 月 2 日向 perplexity/sonar 和 sonar-pro 询问 380 个软件类目的最佳产品，共 760 次调用，收集到 7，534 条引用，其中 59.8% 指向 Tranco 排名 10 万之后、23.4% 指向完全不在百万名内的域名。

ByteDance Seed 等机构发布 HarnessDev 论文（https://arxiv.org/abs/2609.01437），把评估单位从任务输出转为可运行的 harness，分 Creation 与 Evolution 两阶段，按能力和执行 token 成本打分。

普渡大学论文提出 Naive Prompt Optimization（NPO），只保留一条提示词谱系，由教师模型根据近期 rollout 轨迹和奖励迭代修改。在 IFBench 和 HotpotQA 上，NPO 以更少的 rollout 达到与基于 Pareto 选择维护多候选的 GEPA 相当或更好的结果：3，500 vs 3，593 和 6，800 vs 6，871。

Assessing vulnerability detection tools for smart contracts requires datasets with known ground truth, yet such datasets are scarce and difficult to build by hand. We propose an approach that uses Large Language Models (LLMs) to automatica…

Voice actors often re-read the same script while modifying their delivery in response to performance directions. We study this setting as direction-following TTS, where a system generates a new utterance that reflects a given direction rel…

Simulation-based science often requires a distribution over simulator parameters whose push-forward reproduces a set of real observations: this is the source distribution estimation (SDE) problem. Existing methods fit the source against a …

Generative AI is changing how cultural artifacts are created and circulated, and with it our understanding of creativity itself. Researchers disagree about whether these tools enrich or impoverish culture, and we argue that much of that di…