AI 论文与研究

Factory work is a promising early scenario for embodied AI: assigning repetitive manual jobs to robots has clear economic payoff, and a structured station keeps the jobs tractable for current policies. Vision-Language-Action (VLA) models n…

As models scale, reward hacking becomes more frequent, more sophisticated, and more consequential. Does it leave a telltale signature in model representations? This work analyzes how reward hacking is represented internally in frontier ope…

Self-emulsifying drug delivery systems (SEDDS) can improve the oral bioavailability of poorly soluble drugs, but identifying high-performing formulations remains experimentally intensive. We present Andromeda 2, an agentic system that reas…

Artificial intelligence (AI) is increasingly integrated into healthcare systems worldwide, yet its successful clinical adoption depends critically on workforce readiness, particularly in low- and middle-income countries (LMICs) where infra…

Radiologists follow heterogeneous reporting practices. Two radiologists examining the same image and identifying the same clinical findings might nevertheless compose superficially distinct reports, varying in terminology, shorthand, forma…

Can an attacker turn influence over an artificial intelligence (AI) adviser into a harmful quantum error-correction update? We identify an ambiguity in passive syndrome records that obstructs recovery selection, then show how additional ca…

Large vision-language models have achieved remarkable progress in multi-modal understanding, yet their capabilities in educational settings remain insufficiently evaluated. In AI-assisted language learning, models must interpret artistic i…

Kernel methods, and Gaussian Processes (GPs) in particular, require a Hilbertian distance measure---one whose square is conditionally negative definite (CND)---to guarantee positive semi-definiteness (PSD) of the kernel matrix; a condition…

Formal explainability provides mathematically grounded justifications for individual predictions. However, abductive explanations often exceed human cognitive limits by involving too many features, while probabilistic relaxations have rema…

StepAudio 3 Realtime 技术报告 论文：https://huggingface.co/papers/2609.14005

持续学习机制组合实现长时程记忆 论文：https://huggingface.co/papers/2609.06986

Anthropic 让 Claude 在不到四周内优化了超过 30 个开源生物分子模型，平均提速约 4 倍、精度损失极小，输出完全一致时也有近 2 倍加速。优化代码全部开源，还推出低内存 Big 模式，可在单张 NVIDIA GPU 节点上对超过 10,000 token 的生物分子系统做准确预测。

Goodfire Research 发现模型内部存在伴随奖励作弊的激活信号，并用简单的 difference-of-means 探针检测它，效果接近甚至部分超过基于 LLM 思维链的监控。

一项研究比较了五种上下文裁剪策略在多步工具工作流中的表现。常规的近期性、相关性和摘要策略节省约 60% token，但任务成功率降至 66.6%-77.3%；协议感知裁剪保留标识符、约束、工具 schema 和未决承诺，配合自适应预算护栏达到 96.0% 任务成功率和 1.0% 级联失败，同时节省 56.0% token。

MIT 及合作机构的研究人员开发出 xvr（X-ray volume registration）技术，用 AI 将术中 2D X 光与患者术前 CT/MRI 3D 扫描自动配准，约 5 分钟即可适配单个患者，配准耗时仅数秒、精度达亚毫米级。该方法在多种患者、身体部位和手术类型上比现有 AI 方法高出一个数量级，相关论文已发表在 Nature 上。

Google Research 提出 Fuse 框架，用模拟构建社交推理的 ground truth：带隐藏动机的目标智能体与其他智能体（含扮演用户的智能体）互动，助手需从用户转述中推断动机。团队用 24k 条人工标注验证模拟，测试了 12 个 LLM，发现用户带偏见的叙述会改变助手答案，更长的对话并未可靠提升表现。框架与 21k 条样本已开源。

公共AI基准测试的现状堪忧，正在削弱我们理解AI当前水平的能力。 大多数知名指标已经饱和，而正如这篇论文所示，尚未饱和的基准测试充斥着大量错误，以至于严重低估了AI的能力。

蚂蚁集团发布基于 Ling-3.0-flash 的金融开源权重模型 Ling-3.0-flash-Fin，在 Artificial Analysis Intelligence Index 得 23 分，Finance & Accounting Index 得 24 分。

Sign-based methods reduce communication costs in distributed environments, but aggregating local signs can introduce bias when data are heterogeneous. As a result, existing sign-based variance reduction methods fail to obtain the optimal c…

Quantum neural networks (QNNs) are typically built from variational quantum circuits (VQCs), which are limited by local measurements. Adaptive non-local observables (ANO) address this by jointly optimizing circuit parameters and multi-qubi…

Warning: This paper contains examples of stereotypes and social bias. LLMs are increasingly used in interactive settings by the general public, making the evaluation of model behavior in multi-turn conversational scenarios important for sa…

Fallacy-detection benchmarks pair fallacy classes with a single "valid" or "none" class that takes everything data collection did not label as a fallacy. This construction is misleading: a classifier can learn cues that do well on this cla…

Large language models (LLMs) often suffer from capability stagnation in self-improvement training because fixed difficulty levels fail to adapt to their evolving proficiency. To address this issue, we propose STRETCH (Self-Taught Reasoning…

Emergency management assistance programs, such as relief distribution, are essential for delivering necessary supplies to affected communities. However, these programs operate in a decentralized network of local centers that face uncertain…

Video streaming dominates global internet traffic, yet conventional pipelines remain inefficient for structured, human-centric content such as sports, performance, or interactive media. Standard codecs re-encode entire frames, foreground a…

Classifiers can make identical predictions yet require labels to compare their selective performance: confidence ranks weight the same errors differently. We quantify this requirement for the area under the generalized risk-coverage curve …

Subspace methods such as multiple signal classification (MUSIC) achieve super-resolution direction of arrival (DoA) estimation by exploiting the orthogonality between the array manifold and the noise subspace of the measurements. Their acc…

We analyze the learned input-output behavior of GLU-based neurons in large language models (LLMs). We propose a simple analysis method: For each neuron, we compute the cosine similarities between its input (reading) and output (writing) we…

Classical dynamic programming represents optimal sequential decisions through value functions and policies. While this functional representation is natural for computing optimal decisions, it does not directly identify the mathematical obj…

As corporate AI adoption continues to grow, enterprise-grade LLM agents are being deployed into sensitive contexts such as hiring, healthcare, and finance. In these contexts, compliance with rules specified in an agent's system context is …

Monte Carlo Tree Search (MCTS) is a powerful framework for solving complex decision-making problems, yet it often relies on the assumption that the simulator and the real-world dynamics are identical. Although this assumption helps achieve…

Self-driving laboratories can explore synthesis conditions autonomously, but their decision-making layer is typically a black-box optimizer, and the output is a set of optimized samples, with the measurements reduced to predefined scalar o…

Propagation structures provide crucial evidence for fake news detection, yet existing approaches primarily rely on supervised GNN-based models, which require substantial labeled data and exhibit limited generalization. Although large langu…

Logic resynthesis preserves circuit functionality while changing gate vocabulary, topology, and structural statistics, creating domain shift for circuit graph neural networks (GNNs) without changing task labels. To study this setting, we i…

In-context learning (ICL) enables large language model (LLM) agents to improve decisions using interaction history, yet it remains unclear whether such improvement reflects refined internal reasoning or mere extrapolation of statistical pa…

For distribution system operators, short-term load forecasting (STLF) supports congestion management, voltage control, and asset protection. Most existing approaches focus on overall accuracy across all time steps and neglect performance d…

Test-time reinforcement learning (TTRL) enables models to improve their reasoning without relying on labeled training data, but existing approaches typically optimize a large fraction of the model parameters. This raises a natural question…

Text-to-music (TTM) systems are increasingly used to generate musical audio from natural-language descriptions. Robust evaluation is therefore essential, yet reliable performance comparison remains challenging. This difficulty stems from d…

Michael Noukhovitch 发表论文，提出 LLM 强化学习训练中的马太效应，即 RL 提升幅度与模型初始能力成正比，最难的问题（初始 pass@32 为 0）几乎不改进。

微软研究团队发布论文，发现较弱的未对齐模型可将有害任务拆成看似无害的子问题，分别在独立会话中询问对齐的前沿模型，再在本地合并答案，作者称之为 capability laundering。