Anthropic 将每月向埃隆·马斯克的 xAI 支付 12.5 亿美元购买算力

Anthropic 将每月向埃隆·马斯克的 xAI 支付 12.5 亿美元购买算力

3分钟 ·
播放数5
·
评论数0

[00:00] Welcome to Daily Tech briefing. Here are the most important AI and technology stories for today.

欢迎收听每日科技英文速读。下面是今天最值得关注的 AI 与科技新闻。

---

[00:08] OpenAI reasoning model disproves 80-year-old geometry conjecture (猜想). OpenAI claims its new reasoning model has disproved a discrete (离散的) geometry conjecture (猜想) that remained unsolved since 1946. Unlike previous controversial mathematical claims by AI, prominent (杰出的,著名的) mathematicians who initially exposed OpenAI's earlier errors are now backing up this breakthrough.

OpenAI 推理模型推翻已有 80 年历史的几何学猜想。OpenAI 宣称其全新的推理模型成功推翻了一个自 1946 年以来悬而未决的离散几何猜想。与以往引发争议的 AI 数学成果不同,曾指出 OpenAI 早期错误的知名数学家们这次对该突破表示了认可。

---

[00:31] Anthropic to pay Elon Musk’s xAI $1.25 billion monthly for compute. A surprising partnership in the AI industry has been detailed, revealing that Anthropic will pay xAI 1.25 billion dollars per month for computing power. This massive deal highlights the severe scarcity (稀缺,匮乏) of infrastructure (基础设施) required to train next-generation large language models.

Anthropic 将每月向埃隆·马斯克的 xAI 支付 12.5 亿美元购买算力。AI 行业内一项令人意想不到的合作细节被披露,Anthropic 将每月向 xAI 支付 12.5 亿美元以获取计算资源。这一巨额交易突显了训练下一代大语言模型所需的基础设施是多么极度紧缺。

---

[00:54] SpaceX IPO filing reveals xAI lost $6.4 billion last year amid massive expansion. SpaceX's recent IPO filing has offered a rare glimpse into Elon Musk’s AI startup financials (财务状况,财务报表), showing that xAI lost 6.4 billion dollars in 2025. The high burn rate is driven by massive GPU purchases and plans to expand the Grok reasoning cluster (集群,群).

SpaceX 的 IPO 文件披露 xAI 去年因大规模扩张亏损 64 亿美元。SpaceX 最近的 IPO 文件罕见地披露了埃隆·马斯克旗下 AI 初创公司的财务状况,显示 xAI 在 2025 年亏损了 64 亿美元。这一极高的烧钱速度主要由大规模采购 GPU 以及扩张 Grok 推理集群的计划所致。

---

[01:17] Google gears up to remake search with agentic AI at I/O 2026. Google is accelerating its transition (过渡,转型) toward agentic search. According to reports ahead of Google I/O 2026, the tech giant plans to introduce advanced AI agents capable of performing multi-step tasks directly within search, shifting from simple links to action-oriented workflows.

谷歌准备在 I/O 2026 大会上用 Agent 级 AI 重塑网页搜索。谷歌正在加速向 Agent 级搜索转型。根据谷歌 I/O 2026 大会召开前的相关报道,这家科技巨头计划推出能够直接在搜索中执行多步骤任务的高级 AI Agent,从而实现从单纯提供链接向行动导向型工作流的转变。

---

[01:39] Figure AI's humanoid (人形的) robots packaging packages 24/7 goes viral. Figure AI has launched a 24/7 livestream showcasing its humanoid (人形的) robots handling and packaging items. The stream has captured global attention, demonstrating significant interest in the practical deployment of robot labor.

Figure AI 机器人 24 小时打包快递的直播走红网络。Figure AI 推出了一项 24 小时不停歇的直播,展示其人形机器人搬运和包装物品的场景。该直播吸引了全球网民的关注,展示了人们对机器人劳动力实际部署的浓厚兴趣。

---

[02:01] China bans Nvidia’s RTX 5090D V2 chip during Jensen Huang's visit. In a sudden regulatory (监管的) move, China has added Nvidia’s RTX 5090D V2 graphics card to its customs (海关) ban list. This ban took effect while Nvidia CEO Jensen Huang was visiting the country, adding complexity to the company’s regional business strategy.

中国在英伟达 CEO 黄仁勋访华期间禁止销售 RTX 5090D V2 芯片。在一次突如其来的监管行动中,中国已将英伟达的 RTX 5090D V2 显卡列入海关禁售清单。此项禁令生效时,英伟达首席执行官黄仁勋正值访华,这为该公司的区域业务战略增添了复杂性。

---

[02:24] Ramp engineers use OpenAI's Codex with GPT-5.5 for faster code review. Fintech unicorn (独角兽企业) Ramp has integrated OpenAI’s Codex powered by GPT-5.5 into its development cycle. Engineers report that AI assistants now provide substantive review feedback within minutes, significantly speeding up their deployment velocity (速度).

Ramp 工程师使用 OpenAI Codex 与 GPT-5.5 加速代码审查。金融科技独角兽 Ramp 已将基于 GPT-5.5 的 OpenAI Codex 集成到其开发周期中。工程师们反映,AI 助手现在能在几分钟内提供实质性的审查反馈,大幅提升了他们的部署速度。

---

[02:44] That's all for today. See you tomorrow, and thank you for listening.

以上就是今天的内容。感谢你的收听,我们明天见。

---

Vocabulary Notes:

1. conjecture 猜想

2. discrete 离散的

3. prominent 杰出的,著名的

4. scarcity 稀缺,匮乏

5. infrastructure 基础设施

6. financials 财务状况,财务报表

7. cluster 集群,群

8. transition 过渡,转型

9. humanoid 人形的

10. regulatory 监管的

11. customs 海关

12. unicorn 独角兽企业

13. velocity 速度