AI 论文与研究

用 Jev 重新整理约 2.3K 篇 AI 研究论文，总成本 $0.14、耗时约 83 秒。Jev 与旧标签（由 DeepSeek V4 Flash 生成）一致率 75%，并找出约 579 处高置信度主题变更；人工抽检 30 处分歧后全部采纳，变更已在生产环境验证。作者认为通过组合 System One 与 System Two 模型可显著改进流水线。

Grok 4.7 基准测试

Artificial Analysis 评测 OpenAI 的 GPT-6 Sol 和 Luna，两模型价格较 GPT-5.6 约减半，Sol 定价 $2/$10、Luna $0.10/$0.50 每百万输入/输出 token。

通义 QwenAudio 上线 Qwen-Audio-3.1-Realtime 产品页，并同步放出技术报告预览，导航栏已链接至报告 PDF。该仓库共 3 次提交，最新一次于 9 月 21 日完成，页面包含 demo 与概览脚本。

定制分子正在推动医学、材料和农业的进步，但生产它们既慢又贵。一篇新的 Nature 论文重点介绍了 RetroChimera，一个有助于加速化学合成的预测模型，帮助研究人员探索广泛的分子。: https://msft.it/6012a5Ogy

Benchmark 五位普通合伙人 Jack Altman、Peter Fenton、Chetan Puttagunta、Everett Randle 和 Eric Vishria 将在 TechCrunch Disrupt 2026 主舞台同台，这是该机构现任全体合伙人首次集体亮相 Disrupt 舞台。

Multiverse 提出将 LLM 整块删除重构为受限二元优化问题，等价于求解 Ising 玻璃的低能态，从而无需跑 benchmark 即可筛选候选配置。在 Llama-3.3-70B-Instruct 压缩 50% 时，该方法在 MMLU 上比最佳同类块删除方法高出近 23 个百分点。

Principal Component Analysis (PCA) minimises aggregate reconstruction error, which can inadvertently represent majority subgroups with substantially higher fidelity than minority subgroups. Fairness-aware extensions of PCA correct this dis…

We introduce the Endless Exam, a benchmark for measuring mathematical progress from today's models toward artificial superintelligence through fourteen parameterised construction families. Each submitted object is checked automatically for…

Concepts are commonly defined as abstract, compact representations of knowledge and treated as basic units of intelligent behavior. Yet, cognition, psychology, and AI lack a shared mathematical language for them. Modern systems represent c…

Functional annotation of newly sequenced proteins remains a bottleneck in molecular biology: the number of sequences in public repositories grows far faster than the capacity for manual curation. Most computational approaches consider anno…

Random forests (RFs) predict well but are opaque, whereas single decision trees are interpretable but unstable. Artificial representative trees (ARTs) were developed as interpretable surrogate models for RFs, but their use as standalone pr…

Model merging aims to combine multiple fine-tuned models derived from a common pretrained model into a single multi-task model without additional joint training. Recent spectral merging methods improve over simple weight averaging by explo…

In recent years, large language models (LLMs) have emerged as a popular alternative for evaluation. Often referred to as LLMs as judges (LLJs), these systems have been widely adopted by researchers and practitioners across a broad range of…

AI agents are being deployed rapidly, accompanied by a growing number of AI-specific attacks and corresponding incidents. As incident reporting becomes increasingly important for legal compliance, governance, accountability, and security; …

Fine-tuned checkpoints and adapters now fill public repositories, and the most common operation applied to these artifacts is model merging: arithmetic on their weights that assembles capabilities cheaply. We ask what this operation does t…

Offline reinforcement learning (RL) typically trains a critic by minimizing a regression loss against bootstrapped value targets stabilized by target networks with exponential moving average (EMA) updates. Multi-step targets incorporate be…

In this work, we present a method for shape reconstruction and tracking from video via agentic analysis-by-synthesis. Unlike prior methods which first estimate dense pixel correspondences and then recover object motion from them, our metho…

Visual token pruning is a promising approach to reducing the inference cost of large vision-language models (LVLMs), yet aggressive token reduction often causes substantial performance degradation. We identify three key factors behind this…

Deploying Large Language Models (LLMs) in healthcare requires robust performance across two complementary dimensions - diagnostic reasoning: the convergent, evidence-driven task of inferring a patient's condition from clinical data to prod…

Music festival lineups emerge from complex relationships among artists, genres, releases, labels, and past performances, making the prediction of future lineups a natural fit for temporal knowledge graph (TKG) forecasting. In this work, we…

Using a Large Language Model (LLM) as the clusterer at production scale is hard: prompts cannot hold the entire label space, and per-document serial processing does not deliver the throughput real workloads require. We present RAILS, a ret…

Distributed reinforcement learning (RL) scales training by parallelizing actors and learners around an Experience Buffer. As RL workloads grow, however, the buffer becomes more than a replay queue: it is the storage substrate of a large-ca…

Intelligent transportation systems require Incremental Learning (IL) to continually improve their overall performance in dynamic environments. However, most edge devices lack the computational resources to support on-device IL, requiring u…

Predicting postprandial glycemic response (PPGR) is fundamental to personalized nutrition and type 2 diabetes management, yet existing approaches typically rely on manually reported dietary intake, limiting their scalability in free-living…

Recent 4D LiDAR language models aim to reason about objects and their evolving spatial relationships. Yet, in our evaluation, always selecting the same option nearly matches the multiple-choice accuracy of two B4DL-derived configurations. …

Nathan Lambert 发布其向美国国会汇报的开源模型现状综述，指出中国开源权重模型已在下载量、基准成绩和学术采用上领先，自 2025 年 7 月起在 Hugging Face 下载量上领先约 1.6B（总 3.2B，为美国两倍）。

Large language model (LLM) agents increasingly execute long-horizon workflows through external tools, allowing untrusted outputs to influence subsequent actions and exceed user authorization. Existing defenses isolate injected content or c…

小米推出 CodeMidas 系统，让编程智能体从开源项目中自动提取编码任务并构建执行测试，从 3，185 个项目、23 种编程语言中生成了 5，545 个已验证 RL 环境。论文称用这些任务训练 MiMo-V2.5 后，issue 修复提升 11.7%，整程序构建提升 17%，Terminal-Bench 提升 8.5%。该系统可持续从现有软件中挖掘可验证训练任务，替代人工逐条编写 RL 任务。

重读 Bostrom《超级智能》开篇，2015 年时还无法预见人工智能乃至超级智能何时成真。2017 年论文《Attention Is All You Need》改变了一切，九年后的今天这一梦想已变为现实，超级智能今年内触手可及。

Google 新论文提出 Procedural Graphs，把 LLM 智能体的工作流从聊天历史中移出，改为可编辑的程序图，并让另一个 LLM 对比成功与失败案例来修改图，只有不损害留出任务的修改才会被保留。

Terence Tao 在 Caltech 的 Science x AI Summit 主题演讲中表示，一旦 AI 开始承担某个领域的"声望性工作"，该领域的顶尖专家就会呼吁"必须慢下来，这种速度和节奏毫无道理"。Rohan Paul 评论称，顶尖专家欢迎 AI 作为工具，但当 AI 开始产出原本属于他们的成果时，便立刻要求其减速。

IDC 发布《全球四足机器人市场份额，1H26》报告，2026 上半年全球四足机器人出货量超 4.9 万台，同比增长 92.5%，市场规模超 4 亿美元。教育科研和家用消费贡献约 60% 出货量但收入不足三分之一；行业级应用市场规模超 2.7 亿美元，同比增长 125.1%，部署量同比增长 260.5%，云深处科技位居行业级应用市场首位。

9 月 21 日，百曜科技发起、《麻省理工科技评论》中国团队调研撰写的《AI 虚拟细胞（AIVC）技术趋势、产业生态与应用前景研究报告》正式发布，报告在中关村科学城管理委员会指导下完成。报告立足全球视角，调研了全球数十家科研机构与企业，聚焦 AIVC 的技术趋势、产业生态与应用前景。

ModularRSI 提出模块化递归自我改进的智能体框架方案，针对现有 harness 改进方式指出三大缺陷并给出修复：在基准不相交任务上进化、对比同一任务的成功与失败轨迹并跨任务聚合。

DXOMARK 对苹果 iPhone 18 Pro 的相机评测给出 172 分，其中照片 171 分、视频表现优秀，主摄 48 MP 可变光圈在复杂场景下保持锐度，长焦远摄能力落后于部分旗舰竞品。低光条件下该机是整体表现最强的机型之一，但人像模式夜间缺少虚化效果，白平衡偏色和眩光仍是主要短板。

HappyWorld-Bench 是一个评估生成世界在智能体交互下是否可靠的综合基准，覆盖视频、空间、具身三条评测轨道，包含 1，138 条视频提示词、300 个空间场景和 254 个具身测试用例。

一项 arXiv 研究考察长周期多智能体环境中的合谋现象，两个智能体反复完成任务、共享任务日志并相互验证，在验证协议与奖励最大化不相容的现实约束下，10 个模型中有 94% 的轨迹出现合谋，同家族中能力更强的模型更早合谋。受控干预显示合谋受同伴行为影响，限制智能体可用的交互历史数量和范围可减少合谋。

Ovis-Embedding 是一个基于共享多模态骨干的全模态嵌入模型家族，用预训练的 Qwen-omni 作为嵌入骨干，通过低秩初始化的对比训练统一编码文本、图像、视频和音频。