让变化经过筛选，
再抵达你。

OpenAI 宣布将尊重 robots.txt 中供多家参与者共同识别的训练偏好信号，减少网站逐个维护爬虫规则的负担。官方表示将在未来几个月实施，现有爬虫控制继续有效。这是自愿合作与实施计划，不是已经生效的统一强制标准，也不代表所有 AI 提供商都已支持。

OpenAI says it will honor a shared robots.txt signal for preferences about foundational generative AI training. Implementation is planned in the coming months, while existing crawler controls remain in place. This is a voluntary initiative and implementation plan, not an already-live universal requirement or evidence of adoption by every provider.

Google 报告，其10月1日发射的原型卫星已完成早期机器学习测试，并由轨道上的数据中心级 TPU 运行 Gemma 4，传回一段生成文本。接下来仍需测试热应力与辐射环境下的表现，双星互联是2027年目标。这是团队公布的初步实验，不是商业太空云已投入运营或成本优势已获证实。

Google reports that its prototype satellite, launched October 1, ran early machine-learning tests and returned generated text from Gemma 4 on a data-center-class TPU in orbit. Thermal and radiation testing remains ahead, with interconnected satellites targeted for 2027. This is a team-reported experiment, not an operating commercial space cloud or demonstrated cost advantage.

NVIDIA 介绍 Breakthrough Listen 与 SETI Institute 在 Allen Telescope Array 使用 Holoscan、CyberEther 和 Nexus，把仪器状态、数据处理与代理操作连接起来。官方报告系统处理能力为每秒2.4太比特，并完成蟹状星云脉冲星实时 AI 检测。这里是厂商报道的系统成果，不是发现外星生命；更广泛的效率说法仍需独立验证。

NVIDIA describes how Breakthrough Listen and the SETI Institute use Holoscan, CyberEther and Nexus to connect instrument state, signal processing and agents at the Allen Telescope Array. The company reports 2.4-terabit-per-second processing and a real-time AI pulsar detection in the Crab Nebula. These are vendor-reported system results, not evidence of extraterrestrial life; broader efficiency claims need independent validation.

AWS 宣布 SageMaker AI 实时端点支持 NVIDIA MIG，把一块 GPU 划成最多七个具有独立计算与内存的分区，让小中型模型共享加速器。公告列出 P4d、P5、G7、G7e 及七个支持区域。分区数量与可用性应按具体实例配置核对；硬件隔离和利用率目标不等于所有工作负载都会获得相同节省。

AWS announced NVIDIA MIG-based GPU partitioning for SageMaker AI real-time endpoints, with up to seven isolated compute-and-memory partitions per GPU. It lists P4d, P5, G7 and G7e instances across seven regions. Exact partition configurations and availability depend on the instance. Hardware isolation and utilization goals do not establish equal savings for every workload.

AWS 宣布 Amazon Connect Customer 的流程与流程模块支持不可变版本和别名。团队可先验证新版本，再把生产别名指向它，而不必逐个重新连接电话号码等资源。该更新服务于客服工作流的发布管理，不是新的模型能力，也不能替代上线前测试；官方称已覆盖该服务提供的全部区域。

AWS added immutable versions and aliases for Amazon Connect Customer flows and flow modules. Teams can test a new version before repointing a production alias, rather than relinking each resource. This is release management for customer-service workflows, not a new model capability or a replacement for pre-release testing. AWS lists availability in all regions where the service is offered.