让变化经过筛选，
再抵达你。

Google 公告称，用户可用 Alt + Space 调出 Gemini，借助 Gmail、Drive 信息起草摘要，并生成图像。企业使用仍受 Workspace 管理员的生成式 AI 设置控制。这是产品可用性公告，不是独立的效率评测；接入工作资料前，应确认账号权限与组织设置。

Google says Alt + Space opens Gemini for desktop assistance, including summaries drawing on Gmail and Drive and image generation. Workspace administrators retain control through generative-AI settings. This is an availability announcement, not an independent productivity study; account access and organizational settings still matter before connecting work information.

Liberty Global Tech Ventures 宣布投资以存储为优先的 AI 推理软硬件公司 Positron AI，并称参与其 8.75 亿美元融资轮，该轮估值为 50 亿美元。这些金额来自投资方公告。融资体现资本投入，并不能独立证明芯片吞吐量、成本优势或实际交付能力。

Liberty Global Tech Ventures announced an investment in Positron AI, which builds memory-first inference hardware and software. It reports joining an $875 million round valuing the company at $5 billion. These figures come from the investor's announcement. Funding signals capital commitment, not independent proof of throughput, cost advantages, or delivery capacity.

Tempus 宣布未来数年建设包含十万份全基因组的研究平台，关联去标识化临床信息。公司称初始数据已面向早期参与者提供，计划 2027 年中全面开放；百万份基因组是后续目标。这些规模与时间是计划，不是已完成的数据量，也不构成临床疗效证据。

Tempus announced a multiyear research platform targeting 100,000 whole genomes linked to de-identified clinical information. It says initial data is available to early adopters, with general availability planned for mid-2027; one million genomes is a later goal. These are planned scale and timing, not completed dataset counts or evidence of clinical benefit.

Sergio Paniego 在课程配套文章中，把任务与验证器、交互接口、隔离运行环境和训练系统区分开来，并介绍 TRL 与 OpenEnv 的组合。文章说明了为什么持续编辑文件、运行测试的智能体需要保存环境状态。这是作者的教学与报告综述，不是新模型发布或独立性能验证。

In a course companion article, Sergio Paniego separates tasks and verifiers, interaction interfaces, isolated execution, and the training system, and discusses combining TRL with OpenEnv. It explains why agents that edit files and run tests need persistent environment state. This is an educational synthesis of reports, not a new model launch or independent performance validation.

Open Athena 的 Grug 67B A2B SFT 模型卡公开了研究检查点，并强调只做过初步运行测试和探索性检查，未验证生产适用性。标称 262K 上下文是配置值，长上下文能力尚未测试。该资源适合作为研究线索，不宜根据参数名或少量示例推断可靠性。

Open Athena's Grug 67B A2B SFT model card presents a research checkpoint with only smoke tests and exploratory probes, explicitly not validated for production. The advertised 262K context is a configuration value; long-context performance remains untested. It is a research lead, not a reliability claim supported by its name or a few examples.