作者丨 幸丽娟 编辑丨岑 峰  
（翻译）将手语AI交到用户手中
Introducing sign-language-to-text (SL2T), our breakthrough model powering new sign language features for Deaf and hard of hearing users.
（翻译）使用 NVIDIA FLARE 构建联邦多模态 AI 工作流
Modern vision-language models (VLMs) can support tasks such as visual question answering, captioning, and image-text reasoning. In practice, however…
（翻译）CUDA Python 1.0：稳定API、统一基础、完整平台访问
For years, a Python developer who needed a GPU had two realistic choices: Learn NVIDIA CUDA C++ well enough to write an extension, set up a build toolchain…
（翻译）审计合成回忆录：对照其所描述生活的文档记录度量LLM生成自传中的场景级虚构
Abstract page for arXiv paper 2608.23640: Auditing the Synthetic Memoir: Measuring Scene-Level Confabulation in LLM-Generated Autobiography Against the Documented Record of the Life It Describes
（翻译）MolEmb：多模态大语言模型可以成为强大的分子嵌入模型
Abstract page for arXiv paper 2608.23646: MolEmb: Multimodal Large Language Models Can Be Strong Molecular Embedding Models
（翻译）服务掩码扩散大语言模型：基于真实硬件的特性分析与设计原则
Abstract page for arXiv paper 2608.23807: Serving Masked Diffusion LLMs: Characterization and Design Principles from Real Hardware
（翻译）拒绝并非稳健性：在可证明无信息的临床疼痛语音转录中审计大语言模型的自信捏造
Abstract page for arXiv paper 2608.26167: Refusal Is Not Robustness: Auditing Confident Fabrication in Large Language Models on a Provably Uninformative Clinical Pain Speech Transcript
对人来说，推门近乎是一种下意识动作。对移动操作机器人来说，这是一道把感知、移动、操作和控制绑在一起的综合题。 机器人需要找到并靠近把手，完成接触与旋转，在门体移动时协调底盘、机械臂和夹爪，还要避开门框、连续穿过狭窄门口。任何一步出现误差，都可能让整段任务中断。
（翻译）可检验人类知识的时间胶囊：单一免费本地模型中的41年Jeopardy!
Abstract page for arXiv paper 2608.27459: Time Capsule of Testable Human Knowledge: 41 Years of Jeopardy! in a Single Free Local Model
（翻译）挑战性真实场景下通用移动助手的基准评测
Abstract page for arXiv paper 2608.27477: Benchmarking General Mobile Assistants in Challenging Real-World Scenarios
（翻译）思考消耗令牌：更多结构何时物有所值
Abstract page for arXiv paper 2608.27506: Thinking Costs Tokens: When More Structure is Worth the Price
（翻译）使用 NVIDIA Omniverse NuRec 扩展跨车辆平台的自动驾驶感知
A perception stack is shaped by the vehicle that carries it. Move the same software to a new carline—for example, from an SUV to a sedan or another vehicle…
作者丨 幸丽娟 编辑丨岑 峰  
谷歌推出 HEIR（同态加密中间表示）。这是一个开源编译器和开发工具链，能够对通常使用未加密输入构建的预训练 AI 模型进行编译，使其能够处理加密数据。
