Learning more about Claude’s mathematical capabilities

Anthropic 最新研究：Learning more about Claude’s mathematical capabilities…

📅 08-10📎 Anthropic Research

Frontier Red Team

Anthropic 最新研究：Frontier Red Team…

📅 09-10📎 Anthropic Research

GPU-CFR: 80x Faster Counterfactual Regret Minimization by Compiling the Game to Static Dataflow and CUDA Graph Replay

GPU-CFR将博弈树编译为静态数据流，并利用CUDA Graph重放执行，显著降低内核启动与调度开销。在反事实遗憾最小化中实现最高80倍加速，为大规模不完美信息博弈的高效GPU求解提供新方案。

📅 09-10📎 arXiv AI Papers

General Quantification of Covariate and Concept Shifts

针对协变量偏移与概念偏移，提出通用量化框架，能够统一估计两类分布偏移程度，并给出理论保证。方法无需目标域标签即可评估偏移，并在合成与真实数据上验证有效性，为迁移学习与分布鲁棒性分析提供实用工具。

📅 09-10📎 arXiv AI Papers

Can Edge-Deployable Vision-Language Models Identify Species?

评估可部署于边缘设备的视觉语言模型能否识别物种，比较其准确率、推理速度与资源消耗，并与传统模型及云端方案对照。结果揭示轻量视觉语言模型在野外生态监测中的潜力与局限，为实际部署提供依据。

📅 09-10📎 arXiv AI Papers

Generative Marketing Mix Modeling: A Causal Inference Framework Linking GEO and GEM to Business Impact

该研究提出生成式营销组合模型，以因果推断框架将GEO与GEM等生成式营销指标关联至业务成效，量化其对销售和ROI的因果贡献，为企业在生成式搜索与引擎环境中的预算分配和效果评估提供新方法。

📅 09-10📎 arXiv AI Papers

Artificial Id: Drive and Persistent Alignment in Agentic AI

该文提出“人工本我”概念，为智能体AI引入内在驱动力，并设计持续对齐机制，使其在长期自主行动中仍与人类价值保持一致。研究探讨自主性与安全性的平衡，为构建可持久对齐的智能体提供新思路。

📅 09-10📎 arXiv AI Papers

MindTopo: Can Foundation Models Reason in Topological Space?

该研究提出MindTopo，旨在评估基础模型能否在拓扑空间中进行推理。基准涵盖连通性、连续变形与空间关系等任务，系统检验模型的拓扑理解与泛化能力，揭示当前基础模型在拓扑推理上的优势与局限。

📅 09-10📎 arXiv AI Papers

Domain-Specific Hallucination Detection in Large Language Models

Large language models generate fluent text that can contain unfaithful claims — a phenomenon known as hallucination. We present a multi-signal detection pipeline combining fine-tuned DeBERTa-v3 classification, Monte Carlo (MC) Dropout uncertainty quantification, and temperature-scaled calibration f

📅 09-10📎 arXiv AI Papers

Biology-in-the-loop: Amortized Adaptive Hit Discovery in CRISPR Screens

Many biological discovery problems require experiments to be selected sequentially under constrained budgets. CRISPR screening is a prominent example, as exhaustive perturbation testing is often infeasible and candidate perturbations must instead be prioritized over multiple experimental rounds. Des

📅 09-10📎 arXiv AI Papers

On the Regularization Landscape for the Linear Recommendation Models

Recently, a wide range of recommendation algorithms inspired by deep learning techniques have emerged as the performance leaders on several standard recommendation benchmarks. While these algorithms were built on different DL techniques (e.g., dropouts, autoencoder), they have similar performance an

📅 09-10📎 arXiv AI Papers

The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement

本文围绕“人类制造的最后一款AI”构想，探讨实现真正递归自我改进的路径，分析AI自主迭代升级的条件、能力跃迁与潜在风险，并讨论对齐、安全与控制等治理难题，为通用人工智能演进提供前瞻框架。

📅 09-10📎 arXiv AI Papers

ToolGrad: Efficient tool-use dataset generation with textual “gradients”

Machine Intelligence

📅 09-10📎 Google AI Blog

The iPhone Duo, The Intelligent Personal Hub, Apple Watch Audio Intelligence

Ben Thompson 科技战略分析：The iPhone Duo, The Intelligent Personal Hub, Apple Watch Audio Intelligence…

📎 Stratechery