致我们终将逝去的青春

11 high school students arrested over massive brawl in middle of school day_我的网站

一 |     谷歌正在尝试把 Gemini 与芯片更紧密地结合起来。    Eleven students have been arrested following a massive high school brawl that took place in the middle of the school day.The incident occurred at Matanzas High School in Palm Coast, Florida – some 90 miles north of Orlando – when two school resource officers responded to a disturbance in a hallway of the high school at approximately 1 p.m. on Monday afternoon, according to a statement released from Flagler County Sheriff’s Office.All of the schools in Flagler County have at least two school resource deputies assigned to each campus and the two officers “immediately assisted teachers and administrators in attempting to break up the fight and disperse the crowd of on-lookers,” police said.“During the fight, a female student pushed (a deputy) so they could attack another student,” authorities said in their statement describing the chaotic situation. “At the same time, (another deputy) was working to control the situation and observed a male student run into the altercation. Deputy Landi then observed the student charging at a school staff member while swinging his arms, ultimately punching the staff member in the shoulder … At the same time, another school administrator was attempting to escort a student away from the fight, when another male student ran up to the administrator and grabbed them. The student then battered the administrator before being taken to the Dean’s Office.”In total, 11 students were arrested, three of which with felony charges of Resisting a Law Enforcement Officer with Violence and Battery on a School Official. The other eight students had affidavits filed with the State Attorney’s Office recommending charges to be filed against them for misdemeanor offenses, police said.“The lack of respect demonstrated by these students is simply shameful,” said Flagler County Sheriff Rick Staly. “But actions have consequences. Parents, be the Sheriff in your home and teach your kids the importance of respecting teachers, staff, and deputies. Teach them how to handle disagreements and that fighting only leads to more violence. We have a zero-tolerance policy for violence at schools, and you will be arrested. I also commend our School Resource Deputies and the Matanzas High School staff for quickly getting a very volatile situation under control before anyone was seriously hurt.”Flagler Schools Superintendent LaShakia Moore added, “I agree with what our Sheriff has said in the past; this type of behavior is not to be tolerated by anyone. That said, these actions are not indicative of the thousands of students who come to our campuses each and every day to learn. There are so many students, teachers, administrators, and staff who work every day to promote and celebrate a positive school culture at their schools. This is an opportunity for our parents and guardians to reinforce the good choices their children decide to make and remind them why it is important to be successful in life.”Parents have since been notified of the incident by Matanzas High School staff and an investigation into the cause of the altercation is underway.。

二 |          据《The Information》援引知情人士消息，谷歌母公司 Alphabet 正在研发一款服务器 AI 芯片，内部非正式代号为“Frozen v2”。这款芯片针对 Gemini 模型设计，计划最早于 2028 年部署。

三 | 谷歌工程师预计，以单位功耗所能处理的 Token 数量计算，其效率可能达到公司最新一代自研 AI 芯片的 6—10 倍。         谷歌目前尚未正式确认 Frozen v2 项目，也没有公布芯片规格。公司在回应媒体时表示，其团队一直在研究和试验提升性能与效率的新方案，但并非所有项目最终都会进入生产阶段。谷歌同时强调，从底层共同设计硬件和软件，是其打造高度集成系统的重要方式。但报道发布后，Alphabet 股价在当地时间 7 月 20 日早盘上涨约 3.3%。

四 |          与现有 TPU 相比，Frozen v2 的主要区别，是它进一步收窄了芯片的适用范围。

五 | 它的定位是一颗更极致的“专用芯片”。根据现有报道，谷歌计划将 Gemini 的部分架构特征固化在芯片中，针对其计算路径和常用操作进行专门优化。         大模型推理不仅需要完成大量矩阵运算，还要频繁从高带宽内存读取参数。数据在内存与计算单元之间移动，会消耗时间和电力。Frozen v2 希望减少这部分开销，从而提高响应速度，并在相同功耗下处理更多请求。报道所称的 6—10 倍效率提升，具体指每单位功耗能够处理的 Token 数量，并不是芯片峰值算力。         谷歌现有 TPU 同样是面向机器学习的专用处理器，但需要适配不同模型，并承担训练、推理等多种任务。最新一代 Ironwood 单颗芯片配备 192GB 高带宽内存，单位功耗性能约为上一代 Trillium 的两倍，最多可由 9216 颗芯片组成 Superpod。

六 | Frozen v2 不会取代 TPU，而是作为一条更具针对性的产品线，与其并行使用。

七 |          (来源：Google)          但专用芯片也有其劣势，更深的模型与芯片绑定会降低灵活性。芯片从设计到投产通常需要数年，如果此后 Gemini 的基础架构发生较大变化，已经固化的设计可能难以适配。Frozen v2 能否获得长期收益，既取决于实际能效，也取决于谷歌能否在快速迭代模型的同时，维持底层架构的相对稳定。         那么，谷歌为什么要“铤而走险”？首先一个原因，是为了应对持续增长的算力需求。据相关数据，谷歌内部的 AI 计算资源短缺已经引发不同团队之间的资源竞争，Google Cloud 甚至因此拒绝过部分外部客户订单。         大模型的计算需求包括训练与推理两个阶段。训练先进模型需要集中投入大量算力，但随着 Gemini 被加入搜索、办公软件、云服务和智能终端，长期发生的推理可能形成更庞大的成本。

八 | 单次生成的费用看似有限，乘以数以亿计的用户和持续增长的调用量后，电力、芯片、网络及数据中心投入都会成为重要支出。         Alphabet 已将 2026 年资本开支预期上调至 1,800 亿—1,900 亿美元，主要用于扩建 AI 计算能力、数据中心和相关基础设施。在这一背景下，提高每瓦电力能够生成的 Token 数，不仅意味着更低的能源消耗，还意味着同一座数据中心能够承载更多用户请求。

九 | 对于已经拥有大量 Gemini 使用场景的谷歌而言，即使单次推理成本只下降一小部分，规模化之后也可能带来可观节省。         图｜谷歌历年资本开支预期（来源：S&P）          自研芯片还可以加强谷歌对供应链和产品节奏的控制，减少对英伟达等外部芯片供应商的依赖。但这种“减少依赖”并不等于完全替代。Frozen v2 即使顺利投产，预计规模也将低于 TPU；谷歌的数据中心仍会同时使用 TPU、GPU 及其他计算设备。它更像是为高频且相对稳定的 Gemini 任务增加一类成本更低的专用工具。         谷歌也不是唯一走向自研芯片的 AI 公司。2026 年 6 月，OpenAI 与博通公布首款自研推理芯片 Jalapeño。OpenAI 表示，这款芯片围绕大语言模型推理设计，将结合博通的芯片实现和网络技术以及 Celestica 的服务器系统能力，计划于 2026 年底开始部署。         图｜博通公司首席执行官 Hock Tan 将一片芯片晶圆递交给 OpenAI 公司首席执行官 Sam Altman。（来源：OpenAI）          Anthropic 也被曝已经开始研究自有 AI 芯片，并与三星电子等潜在合作方接触。不过该项目仍处于早期阶段，芯片的用途、性能目标和服务器部署方式尚未明确 Anthropic 同时表示，亚马逊 Trainium、谷歌 TPU 和英伟达 GPU 仍将是其扩张算力的重要基础。中国方面，此前 DeepSeek 和近期的月之暗面也有自研芯片方面的传闻。         除专注于基础模型的公司外，亚马逊、微软、Meta、阿里巴巴、百度和字节跳动等拥有云服务或大规模互联网业务的企业，也在不同程度上推进专用 AI 芯片。其中部分芯片服务于模型训练，部分集中于推理、推荐和广告等高频工作负载。它们共同反映出一个趋势：随着 AI 服务规模扩大，掌握模型的公司越来越希望同时参与芯片和计算系统设计。         这些项目属于同一轮趋势：AI 公司希望根据自身模型和业务负载设计芯片，以获得更好的成本、能效和供应保障。

十 | 但它们并不是完全相同的技术路线。Trainium、Maia、MTIA 以及 OpenAI 的 Jalapeño 主要面向一类或多类 AI 工作负载，仍保留一定通用性；Frozen v2 则称会把 Gemini 的部分架构直接固化进硬件，模型与芯片的绑定程度可能更高。         因此，Frozen v2 所代表的不只是谷歌扩大自研芯片投入，也是一项更激进的软硬件协同试验。答案将取决于芯片能否如期投产，也取决于到 2028 年时，Gemini 是否仍然适合这块为它提前设计的硅片。

十一 |          参考链接：          1.https://techcrunch.com/2026/07/20/google-is-working-on-a-new-ai-chip-designed-to-make-gemini-more-efficient/          2.https://www.reuters.com/business/google-plans-new-chip-run-gemini-models-more-efficiently-information-reports-2026-07-20/          3.https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/ironwood-tpu-age-of-inference/          4.https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/ironwood-google-tpu-things-to-know/          运营/排版：何晨龙          注：封面由 AI 辅助生成。

Current article:http://kee.zhuaicaisaanzhenshunnaiwei.bond/list_fji4/svoxin.html

Published on:11:15:11