
追踪
────────────────────
[01]OpenAI
- 进展描述
[02]AI Agent 2026
- 行业共识:AI Agent时代已全面到来。英伟达CEO黄仁勋、高通CEO安蒙等科技巨头在COMPUTEX 2026展会上相继宣布AI Agent的时代到来,产业链企业全面跟进。
- AI Agent核心能力演进:从被动对话向主动执行转变。AI Agent具备自主规划、多步骤任务执行、外部工具调用等能力,正在从聊天机器人升级为数字员工。
模型发布
────────────────────
[01]Introducing new capabilities to GPT-Rosalind
- GPT-Rosalind增强生命科学研究能力,提供生物推理、药物化学、基因组学分析和实验工作流程支持。
[02]How Wasmer used Codex to build a Node.js runtime for the edge
- Wasmer用Codex和GPT-5.5构建边缘计算Node.js运行时,开发效率提升10-20倍,交付周期从月级压缩到周级。
[03]A blueprint for democratic governance of frontier AI
- OpenAI提出美国前沿AI治理框架,涵盖安全、韧性和国家安全的联邦层面规范。
[04]OpenAI public policy agenda
- OpenAI发布AI公共政策议程,涉及安全、青少年保护、劳动力转型和全球标准制定。
工具
────────────────────
[01]Direct Preference Optimization Beyond Chatbots
https://huggingface.co/blog/Dharma-AI/direct-preference-optimization-beyond-chatbots
- 直接偏好优化技术扩展超越对话机器人,适用于更广泛的LLM应用场景和模型对齐。
研究
────────────────────
[01]Visual Graph Scaffolds for Structural Reasoning in Large Language Models
- 研究图结构对LLM结构推理的增强作用,探索图作为内部推理机制而非仅外部知识源的价值。
[02]AURA: Action-Gated Memory for Robot Policies at Constant VRAM
- 提出机器人策略的动作门控内存机制,在固定显存下优化长序列单任务推理,适配机器人而非数据中心场景。
[03]Evaluating Transformer and LSTM Frameworks for Prediction in Ungauged Basins
- 对比Transformer和LSTM在无测量流域水文预报中的性能,处理观测缺失导致的高不确定性。
[04]BehaviorBench: Modeling Real-World User Decisions from Behavioral Traces
- 构建真实用户决策基准数据集,用真实行为数据而非模拟用户,支持个性化决策支持系统评估。
[05]ChatHealthAI: Aligning Electronic Health Record Representations with Large Language Models for Grounded Clinical Reasoning
- 对齐EHR表示与LLM能力,融合结构化医疗纵向数据与自然语言推理用于临床决策支持。
[06]Human-in-the-Loop Contextual Bandits for Short-Term Rental Dynamic Pricing
- 短租动态定价中的人机交互式上下文赌博机算法,权衡财务风险、可解释性与稀疏反馈。
[07]Spectral Asymptotics of Neural Network Loss Landscapes
- 精确分解神经网络损失曲面的曲率指数,揭示不同层类型的Hessian特征值尺度规律。
[08]Making Brain-Computer Interfaces More Secure
- 基于脑机接口安全性研究,重点关注EEG机器学习分类器的对抗鲁棒性与隐私保护。
[09]Assessing Region-Level EEG Contributions to Cognitive Workload Prediction
- 评估脑区EEG信号对认知负荷估计的贡献度,提升跨域泛化能力用于安全关键系统。
[10]Testing the Test: Score-Direction Instability in Class-Split Anomaly Detection
- 揭示类别分割评估协议在异常检测中的问题,指出表示空间重叠导致的评估不稳定性。
行业动态
────────────────────
[01]How virtual power plants could provide energy for data centers
https://www.technologyreview.com/2026/06/03/1138350/virtual-power-plants-data-centers/
- Google与虚拟电厂Voltus达成协议,通过需求响应为数据中心供能,优化电网资源配置。
[02]The Download: Trump's new AI order, and smart glasses for warfare
https://www.technologyreview.com/2026/06/03/1138322/the-download-trump-ai-order-smart-glasses-warfare/
- 特朗普签署新AI行政令,涉及人工智能发展战略与国防应用,包括军事级智能眼镜。
[03]Google's new open source Gemma 4 12B analyzes audio, video
https://venturebeat.com/technology/googles-new-open-source-gemma-4-12b-analyzes-audio-video-and-runs-entirely-locally-on-a-typical-16gb-enterprise-laptop
- Google开源Gemma 4 12B多模态模型,支持音视频分析,可完全本地运行于16GB企业笔记本。
[04]Enterprise AI agents keep creating data silos. Microsoft's Build answer is Microsoft IQ and Rayfin.
https://venturebeat.com/data/enterprise-ai-agents-keep-creating-data-silos-microsofts-build-answer-is-microsoft-iq-and-rayfin
- Microsoft推出Microsoft IQ和Rayfin解决AI Agent导致的数据孤岛问题,统一企业数据与业务规则。
[05]Lovable signs multiyear deal with Google Cloud to up usage 5x
https://techcrunch.com/2026/06/03/lovable-signs-multi-year-deal-with-google-cloud-to-up-usage-5x-source-says/
- Lovable与Google Cloud达成多年协议,使用量扩张5倍,获得Claude模型更深度集成。
[06]Alphabet's record-breaking $85B raise for Google's AI business
https://techcrunch.com/2026/06/03/alphabets-record-breaking-85b-raise-for-googles-ai-business-is-a-helluva-good-signal/
- Alphabet创纪录融资850亿美元用于Google AI业务,显示投资者对AI相关产业的强劲需求。
[07]Google's Dreambeans, its weirdest-named AI tool to date, will turn your life into a cartoon
https://techcrunch.com/2026/06/03/googles-dreambeans-its-weirdest-named-ai-tool-to-date-will-turn-your-life-into-a-cartoon/
- Google推出Dreambeans工具,利用用户Google账户数据生成个性化AI插画故事集。
[08]Amazon will show AI product images when you search
https://techcrunch.com/2026/06/03/amazon-will-show-ai-product-images-when-you-search-for-some-reason/
- Amazon视觉搜索集成AI生成商品图片,根据搜索查询展示AI匹配商品,引导用户发现产品。
[09]These two founders left Goldman and Meta to build voice AI for markets everyone else overlooked
https://techcrunch.com/2026/06/03/these-two-founders-left-goldman-and-meta-to-build-voice-ai-for-markets-everyone-else-overlooked/
- 前Goldman和Meta员工创业,为非洲中东市场开发语音AI,日处理超17000通电话。
[10]OpenAI and Anthropic Sign Letter to Prevent AI-Developed Biological Weapons
https://www.wired.com/story/openai-anthropic-letter-ai-biological-weapons/
- OpenAI和Anthropic等企业向议员呼吁加强合成DNA追踪,防止AI被用于研发生物武器。
[11]xAI Asks Court to Strip Alleged Grok Deepfake Nudes Victims of Anonymity
https://www.wired.com/story/xai-asks-court-to-strip-alleged-grok-deepfake-nudes-victims-of-anonymity/
- xAI要求法院取消Grok深度伪造受害者匿名身份,诉讼者或需公开真名或撤回诉讼。
[12]The Humanoid Robot of the Future Is a 6-Foot-Tall Beefcake With a Chinese Body and an American Brain
https://www.wired.com/story/nvidia-unitree-humanoid-robot-h2-plus/
- Nvidia新型人形机器人整合中国制造身体与美国智能大脑,代表国际协作的机器人发展方向。
[13]This Is How Trump Finally Signed the AI Executive Order
https://www.wired.com/story/this-is-how-trump-finally-signed-the-ai-executive-order/
- 特朗普在推迟一月后正式签署AI行政令,标志美国AI监管政策框架的最新进展。
[14]Nvidia's RTX Spark Laptops Look Hell-Bent on Disruption
https://www.wired.com/story/nvidia-rtx-spark-laptop-disruption/
- Nvidia RTX Spark芯片有望真正实现AI PC概念,芯片设计旨在破坏传统计算机市场格局。
付费速览
────────────────────
[01]Claude Code (参考)
- 计费模式:按用量计费(Usage-based)
- 方案说明:通过 Anthropic API 计费,无独立订阅,按 token 消耗付费
- 价格详情:Claude Sonnet 4: $3/M input, $15/M output;Claude Opus 4: $15/M input, $75/M output
- 官网:查看最新定价
[02]Cursor (参考)
- 计费模式:订阅制 + 用量计费
- 方案说明:Hobby 免费(2000次/月补全);Pro $20/月(无限补全 + 500次高级请求);Business $40/用户/月
- 价格详情:高级模型(GPT-4o、Claude)按请求计费,超出套餐后按 token 收费
- 官网:查看最新定价
[03]Trae (实时)
- 计费模式:免费 + 订阅制
- 方案说明:根据页面内容,Trae 的最新付费方案如下:
- Lite $3/月(月付);Pro $10/月(首月免费试用7天);Pro+ $30/月;Ultra $100/月。各档次包含不同额度的基础用量、无限自动补全和云任务并发数(2-20个)。按月计费,支持多数付款方式。
- 价格详情:字节跳动出品,国内版免费额度较高,海外版定价参考官网
- 官网:查看最新定价
夜雨聆风