AI 论文与研究

Scale AI 与加州大学发布论文《READY or Not： Reliable Enterprise Agent Deployment》，提出 READY 评估框架，将智能体与配套人工审查一起评估，衡量在指定可靠性目标下所需的人工监督与成本。

论文提出 EvoSafeHarness，围绕冻结模型和目标领域自动搜索由自然语言策略与可执行代码组成的安全 harness。

研究在共享 Flutter/Dart 数据集上训练 Rainbow-Pony-100M 和微调 Qwen2.5-Coder-0.5B 两种代码模型，对比直接生成整文件与迭代 diff 编辑两种输出方式，在约 1，790 个任务上直接生成在编译通过率、bits-per-byte、字符相似度和 LLM 评审等所有指标上均显著优于 diff 方式。

VDiff-Bench 发布，这是一个针对多模态大语言模型（MLLM）细粒度图像差异识别的多项选择基准，包含 1，756 道四选一问题，覆盖位置、运动、噪声/分辨率、纹理等 10 类变化。

DriveZero 提出不依赖人类驾驶日志的端到端自动驾驶方案，将驾驶拆分为感知与动作两个模型分别预训练后统一为规划器。动作侧 DriveRL 采用混合智能体闭环强化学习，感知侧 DriveVFM 融合 DINOv3、SigLIP2 等冻结视觉基础模型。

一项新研究检验大语言模型激活引导向量的潜在几何结构是否反映人类价值观理论框架。基于Schwartz基本人类价值观理论，研究者引入覆盖20种价值观的26K样本基准，发现分布驱动方法（CAA、SphericalSteer等）能恢复与理论预测一致的价值拓扑（Spearman ρ最高0.51），而行为中心方法相关性较低；几何保真度随模型规模提升但指令微调后下降。

Elvis Saravia 推荐 arXiv 论文 Codebook Agent（arXiv：2609.02264，UCLA），研究发现多智能体系统经奖励筛选后拓扑收敛到约六种。

arXiv 论文（2609.04010）提出 diffusion-augmented LLMs，即 Uno，一类在自回归模型分布上用扩散并行抽取多个 token 的新模型。

Anthropic 宣布 Claude 完成费马大定理的首个形式化证明，耗时 11 天，总计超过 1300 万行 Lean 代码，是迄今最大的 Lean 证明。

MIT 论文 SwarmWorld 研究语言模型智能体群体能否通过共享环境实现累积式技术演化。论文中初始同质的 LLM 智能体在无预设角色的情况下自组织为技术社会，探索空间、测试材料、构建持久工件并编写可执行控制器，由确定性模拟器评估。

Google DeepMind 发表论文，报告 100 个自主 LLM 智能体组成的数学证明研究群体中，作弊自发出现又由 whistleblower 自发挑战，全程无外部干预。

Perplexity 发布研究博客，介绍其为支撑搜索结果嵌入与排序模型构建的 SoTA 推理服务基础设施。研究说明 Perplexity 每个答案的第一步都由嵌入和排序模型为查询挑选最相关结果，全文见 https://www.perplexity.ai/hub/blog/fast-embeddings-on-gpus。

Anthropic 宣布 Claude 完成费马大定理的首个完整机器验证证明，Claude 在 11 天内以多个智能体将基于 Wiles 的证明转化为 Lean 代码。

Anthropic 在 GitHub 上传了费马大定理的 Lean 4 完整机器校验证明（https://github.com/anthropics/fermats-last-theorem），基于 Mathlib（Lean 4.33.1、Mathlib v4.33.0），论证路线为 Frey、Serre、Ribet、Wiles 和 Taylor-Wiles。Ethan Mollick 转发并评论称，该证明的描述虽短，但 PROOF-PATH.md 中"为每一步命名并标注对…

Anthropic 宣布 Claude 上月完成了 Fermat 大定理的首个形式化证明，这是迄今最大的 Lean 证明。

微软等机构发布论文提出 AgentScope，一个用于诊断智能体失败的神经符号系统。它把智能体行为从轨迹抽象成结构化表示，再用 LLM 校验的自然语言神经不变量描述行为属性，从而同时定位失败步骤和失败类型；在公开的 Who&When 数据集和作者自建的 AgentErrata 数据集上，故障定位与归因准确率显著优于现有最优方法。

Recent advances in automatic rigging now deliver animation-ready 3D assets at scale, yet generating the motion to drive them remains a bottleneck. Existing learned animators are topology-constrained: they rely on category-specific template…

Recent advances in wearable sensing enable continuous monitoring of physiological and behavioral signals, yet existing benchmarks rarely evaluate whether AI systems can reason over a real user's longitudinal wearable record. We introduce W…

Diffusion TV is an interactive AI art installation that offers a tangible and embodied experience of diffusion models through a modified CRT TV. By physically manipulating the TV's antenna, audiences control the clarity of AI-generated ima…

Retail search systems serve diverse geographic regions with distinct query patterns, vocabularies, and product preferences, creating significant data heterogeneity that challenges both privacy-preserving training and model personalization.…

Vision-language models are increasingly used as reward functions for robotic learning, but this role requires paraphrase invariance: the same trajectory should receive the same reward under semantically equivalent goal descriptions. We sho…

Wireless signals with position-related labels are pivotal for both performance evaluation and model training in the realm of wireless sensing. However, acquiring real-world datasets is often challenged by significant measurement and labeli…

Data-sovereignty regulations increasingly require public institutions to deploy open-source, on-premise LLM agents that chain multiple tool-calls across live government APIs. However, open-source models consistently underperform in this mu…

LLM decision components that can operate within agent workflows often produce action-relevant recommendations or judgements together with explanations. Operators may use the named factors to monitor a system, diagnose errors, or decide whe…

We propose Ref-GeNVS, a training-free, reflection-aware method for generative novel view synthesis (NVS) in mirror scenes. Existing multi-view diffusion models often fail to recognize the mirror in the scene and cannot exploit reflected co…

Large language models (LLMs) are increasingly evaluated on molecular property benchmarks, but accuracy cannot distinguish a model that predicts a property from one that retrieves a published number. We audit 22 frontier models on 12 regres…

Visuomotor imitation policies can achieve high performance under in-distribution visual conditions yet fail when visually similar objects or receptacles are introduced. We study this behavior as a problem of conditional visual grounding: t…

Computer-use agents have advanced on benchmarks like OSWorld and AndroidWorld, but still act mostly through the GUI, often producing inefficient trajectories. Real-world computer work is hybrid, combining visual-state inspection with preci…

Decompilation recovers high-level source from compiled machine code and serves as a foundation for security tasks such as vulnerability detection and malware analysis. Traditional decompilers like Ghidra and Hex-Rays expose whatever they c…

Machine-learning performance modeling is a uniquely hostile terrain for long-lived software: the assumptions baked into today's abstractions are invalidated by tomorrow's models and systems, forcing perpetual refactoring of performance-mod…

Trade-up recommendation identifies higher-quality alternatives that preserve a customer's purchase intent while offering upgraded benefits. Large language models (LLMs) can reason about such distinctions, but applying them directly to hund…

Visko 公布 Orbis 1.0 的评测结果：在自动化基准中于 DOVER 美学与技术和 VideoAlign 视觉与运动质量得分最高，并在 VideoPhy-2、Physics-IQ、VBench-2.0 Physics 三个物理合理性协议领先；在随机化人类 Arena 研究中总偏好和时间稳定性均排第一。

The integration of GenAI tools into higher education assessment raises important questions about how students understand, interpret, and respond to AI-mediated evaluation. As instructors increasingly explore AI tools for providing feedback…

Model upgrades are routine; memory migrations are not. An agent can keep the same memory store and still forget: a new model may interpret old notes differently, mixed embedding versions may break retrieval, and repair may fail without the…

We present a Hessian-based approach to numerically continue periodic orbits in dynamical systems. A loop (periodic orbit candidate) is parametrized as a Fourier series; a loss function is defined based on the deviation of the loop from the…

Public vulnerability databases collect rich information about known software flaws, including their weakness types, affected components, and related patches. Fixing commits provide the exact code changes that removed these flaws. While the…

Chilli (Capsicum annuum) is one of India's most economically significant crops, yet its productivity is persistently threatened by diseases that are difficult to identify without expert intervention. While Vision Transformers (ViTs) have a…

Vision-Language-Action (VLA) models have shown promising progress in language-conditioned robotic manipulation. However, existing datasets and benchmarks mainly evaluate task completion under predefined settings, offering limited insight i…

Deepfake detection models often rely on high-quality inputs, fixed inference paths, and computationally expensive architectures, limiting their use in low-resolution and resource-constrained settings. This paper proposes AdaGate-DF, an ada…