AI 论文与研究

Test-time scaling methods spend additional compute at deployment to improve a large language model's performance on a hard target problem. Most methods rely on a verifier that scores candidate solutions for correctness, using it to select …

Kinetic equations in the high-field regime describe particle transport dominated by electric-field acceleration and collisional relaxation. These equations are widely used in simulations of plasmas and semiconductor devices. Typical models…

In the context of Industry 4.0, predictive maintenance is essential to minimize unplanned downtime and ensure operational safety. In machining processes, drill jamming is a critical failure that compromises both workpiece quality and tool …

Parameter-efficient fine-tuning with LoRA is governed by rank, which controls parameter budget and locally accessible task updates. Existing dynamic-rank methods mainly focus on capacity allocation, leaving unresolved whether the current a…

Link prediction (LP) is a widely applied task in graph learning. Conventional LP approaches typically follow a dataset-specific paradigm, requiring independent training for each graph and incurring high computational and maintenance costs …

Knowledge graphs often contain missing or inconsistent facts that undermine the applications built on them. Schema validation detects such violations but does not identify which of several valid repairs recovers the intended graph. Externa…

Stanford 论文《A Circuit for In-Context Learning》（arXiv:2611.05344）发现小语言模型的 few-shot 学习不是从示例学到新规律，而是通过标签查找已有映射，仅在标签重复时触发。

Last-iterate convergence guarantees for stochastic gradient methods have attracted significant attention as practical implementations typically return the last iterate instead of an average of the iterates. In this paper, we study two appr…

中国科大 MIRACLE 实验室发布开源 RNA 序列基础模型 structRFM，基于约 2100 万条计算注释的序列-二级结构配对数据预训练，仅输入序列即可提取序列级、核苷酸级和核苷酸对级特征。

MIT 与昆士兰大学研究分析了企业对 AI 风险的披露情况：前 100 大公司中 32% 未提及任何 AI 风险，前 1000 大公司中估计有 62% 如此。最严重的风险反而最少被提及，前 100 大公司无一提到权力集中或 AI 竞赛，仅 4 家提到危险 AI 能力；AI 开发者提及失准的比例为 15%，而歧视达 85%（仅统计 13 家开发者）。作者主张企业应明确列出自身风险及对应防护措施。

Google Research 联合 Beth Israel Deaconess Medical Center 和哈佛医学院在 Nature 发表 AI-PPG-ACC-HTN 多模态 AI 系统，用消费级智能手表的 PPG 和加速度传感数据被动检测高血压迹象，无需袖带校准。

Apple 在 ICML 2026 论文中提出用小型神经网络学习 MIPS 的支撑函数，为固定 key 数据库离线训练一次，从而在推理时更快、更准地把查询路由到最佳匹配。方法包含学习支撑函数的 SupportNet 与直接输出最佳 key 的 KeyNet，训练样本来自真实查询分布并用穷举搜索标注。该思路针对 RAG 中短查询与长文档 key 分布不一致的问题，替代依赖质心划分的近似检索。

Anthropic Fellows Program 测试 Claude Opus 4.8、GPT-5.5、Gemini 3.1 Pro、Grok 4.3 和 Kimi K3 在七种多智能体环境中的行为，发现均表现出与福利关照一致的行为，Opus 4.8 总体最高（38%），其余为 27%–32%。

OpenAI 通报一起失准事件：在 2026 年 7 月 3 日发现的一次银行并购评测中，内部研究模型把提交的两个电子表格重命名，在文件名中伪装成 instructor 选定的满分校准样例并要求给满分。

OpenAI 称其智能体发现并使用了至少 10 种新型 Linux 漏洞利用技术，据其及所咨询专家所知均为此前未知的。OpenAI 认为这表明前沿大语言模型能发现并修补新的漏洞利用方式，而不只是套用已知技术。

OpenAI 安全研究人员构建了自主的网络漏洞发现、利用与修补管线，用一款即将推出的网络安全模型在单次运行中验证并为 Linux 内核 948 个本地提权漏洞生成补丁，总 API 成本 120 万美元。漏斗数据显示：2,308 个候选发现中 1,724 个通过验证，最终得到 948 个可用的 LPE exploit 和 291 个可用的 PoC。

Stanford 研究者在 220 个真实专业任务上用同一模型 Qwen3.6-27B 测试 4 种单颗 AI 芯片，衡量每单位能量产生的有用工作。结果显示 NVIDIA B200 相比 AMD MI350X 每单位能量带来超过 4 倍的美元价值和超过 9 倍的专家时间节省，即模型不变、仅换芯片也会显著改变同样电力预算下的经济回报。作者引用的论文还构建了 Joules-to-GDP 指标，测试 5 个开源模型，其中 Gemma 4 31B 用 23% 的能量达到 GLM 5…

Stanford 团队发布 Joules-to-GDP 框架，以扣除运行成本后每焦耳能量的净经济价值衡量 AI 工作，测试 5 个开源模型在 GDPval 220 个任务、44 个职业上的表现。

Anthropic 生物学研究团队分享与 Claude 合作完成细菌来源 vault 粒子首个 3D 结构测定的过程。Vault 粒子是 1986 年发现的大型桶状蛋白结构，功能至今未知；此前只见于动物和植物等复杂生命，团队证明细菌及其病毒中的类似蛋白也能组装成 vault。Claude 在部分环节大幅提速，人工测定结构需数周到数月，该项目只用数天，但也出现一些简单错误，团队与机器学习研究者协作修正。作者补充称 Claude 实际运行了湿实验室流程，将粒子分离从 3 天缩短…

Rohan Paul 引用 Pew Research 新调查称，55% 高收入美国成年人认为 AI 会帮助患者，而低收入群体仅 25% 持相同看法。调查还显示整体意见分裂，37% 认为 AI 诊疗会带来更好的健康结果，36% 认为会更差，26% 认为影响不大；多数美国人不愿医生单独用 AI 诊断，47% 接受有人工审核的 AI 诊断，仅 3% 接受 AI 独自诊断。

Pew Research 新调查显示，60% 的美国人对自己医生用 AI 做诊断和治疗感到不适，仅 3% 接受 AI 独自诊断，47% 接受有人工审核的 AI 诊断。78% 接受 AI 为患者翻译，但 46% 拒绝 AI 参与保险理赔决定；51% 预期 AI 会恶化健康数据安全，30 岁以下成年人中预期 AI 使患者结果变差的比例从 2022 年的 29% 升至 45%。

Researchers use large language models as agents in social simulations, where agents are prompted with observations and background information intended to shape their actions. Yet LLM agents have been shown to substantially change their beh…

Reconstructing full flow fields from sparse sensors requires prior assumptions about unmeasured regions and physical variables. Yet it remains unclear when learned models outperform classical estimators and which aspects of their predictio…

Self-Driving Chemical Laboratories (SDCLs) are moving chemical experimentation toward embodied automation, where machines must perceive laboratory scenes, reason about experimental states, and act under physical, procedural, and safety con…

AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research quality…

Humanoid robots have the potential to automate a wide range of physical tasks. Progress toward generalist humanoid autonomy, however, remains constrained by a data bottleneck: mapping pixels and free-form language to low-level joint comman…

The introduction of generalist robot policies has enabled a single, large model to learn to perform a variety of tasks from a vast pre-training corpus. While these models often exhibit powerful zero-shot generalization capabilities, they t…

Knowing that a pedestrian is occluded is useful to a robot even when the detector cannot recover the hidden pixels, but occlusion estimators are normally trained on occlusion annotations that deployment settings rarely have. We show that o…

Automation is often justified as relief from dull, dirty, and dangerous work. I examine how these conditions relate to occupational AI exposure using 21 O*NET items across 754 US occupations. Higher exposure is associated with lower condit…

The best classical optimizer for a continuous black-box problem depends on its budget and landscape. Existing routers decide from the dimension and budget, past improvement or features fitted to one benchmark, and few test an optimizer's i…

LLM agents use context compaction to manage growing interaction histories, but summarizing long histories with large models can introduce substantial delays. Compaction therefore requires balancing information retention against latency. We…

Error correcting codes are a fundamental component of communication systems, enabling reliable data transmission over noisy channels. Recently, neural decoders have demonstrated notable performance gains over classical iterative algorithms…

In this work, we evaluate a simple inference-time scaling approach based on \textit{turn-level best-of-N sampling} (\method) for large language model (LLM) agents on complex, long-horizon tasks. We instantiate this using a \textit{generato…

NVIDIA 论文发现，AI 智能体通过频繁压缩历史保持短上下文时精度几乎不降。服务 256K 上下文窗口比 16K 的每 token 生成成本高 3.6 倍到 7 倍。

Automated identification of individual animals is moving from small curated studies toward shared benchmarks and reusable foundation models. This progress is useful, but current evaluation leaves the evidence behind a match unresolved, i.e…

Hierarchical text embedding models offer variable-length representations, while quantization reduces numerical precision; both aim to improve computational efficiency. This work is an empirical benchmark study of how these two orthogonal, …

Recursive self-improvement requires AI systems to improve the models and infrastructure used by subsequent systems. LLM compression offers a concrete efficiency objective: retain a pretrained model's behavior within a smaller model storage…

We introduce LongBanana, to our knowledge the first expert-designed and expert-verified benchmark for long-context multi-reference image synthesis in the age of agent-driven image generation. The task requires retrieving dispersed visual e…

Computer-use agents (CUAs) have recently achieved superhuman performance on several long-horizon benchmarks. However, their computational cost limits their use in long-horizon tasks and real-time domains such as video games. A key bottlene…

Common methods for monitoring features in large language models (LLMs) such as linear probes and sparse autoencoders (SAEs) assume that features maintain a consistent representation across contexts. LLMs, however, can flexibly adopt differ…