琅琊榜

IT之家 8 月 22 日消息，据彭博社报道，Anthropic 于周五透露，前谷歌 TPU 项目创始高管之一阿米尔 · 萨莱克（Amir Salek）将加入 Anthropic 的算力团队。

B | 在 2022 年离职前，萨莱克曾长期负责谷歌的 TPU 业务，主导并交付了前七代 TPU 芯片。萨莱克还曾就职于英伟达，在离开谷歌后，萨莱克曾出任私募股权投资机构塞伯罗斯资本管理公司的高级董事总经理。目前，Anthropic 的芯片主要采购自英伟达、谷歌以及亚马逊等多家供应商。

C | 不过，这家总部位于旧金山的公司近期已释放出打造内部自研芯片业务的信号，并已启动相关岗位的招聘。

D |          (ECNS) -- Hot days worldwide could experience around four additional hours of extreme heat by the end of the 21st century under a high-emissions scenario, according to a study led by Chinese researchers.    Published in Nature Climate Change, the study found that traditional daily temperature indicators may overlook shorter periods of extreme heat, while hourly temperature analysis can provide a more detailed picture of heat exposure and associated risks.        Researchers from Sun Yat-sen University and other institutions analyzed global hourly temperature data from 1981 to 2023, combining observational records with climate model projections to assess how extreme heat exposure could change in the future.    The research team found that the duration of extreme heat increased by 62 hours per decade over the study period. Around 28% of extreme heat events occurred on days that would not have been classified as hot days based on conventional daily temperature indicators.    Under a high-emissions scenario known as SSP5-8.5, the researchers projected that between 2070 and 2099, each hot day could experience around four additional hours of extreme heat compared with the 2001-2023 period. Days not classified as hot could also see increased periods of extreme heat.    The study also found that low- and middle-income countries account for a disproportionately large share of both current and projected extreme heat exposure. Younger generations in low-income countries are expected to experience the greatest increase in lifetime exposure.    Researchers said the findings could help improve heat-warning systems, strengthen assessments of health risks and support climate adaptation efforts.    (By Helen Mo, intern Liu Shuangjing)                    。

E | IT之家获悉，与其他 AI 巨头类似，Anthropic 正全力扩充数据中心基础设施以支撑其发展野心。

F | 自研定制芯片不仅有望缓解供应链紧缺的压力，还能通过量身定制的硬件架构更好地匹配自身算法需求。其竞争对手、ChatGPT 的母公司 OpenAI 也在布局相同赛道。OpenAI 此前已公布了与博通联合研发的自研芯片“Jalapeño”，并计划于今年晚些时候正式投入使用。