![[a9072c08-cf3d-43bd-9319-e16b8a53dd3f.mp3]]

Meta酝酿超百亿美元AI投资并洽谈入股Scale AI

简报：

• 据知情人士透露，Meta正与潜在合作伙伴就逾100亿美元的大规模AI投资进行谈判，旨在加快其在生成式AI和大语言模型等领域的研发和基础设施建设。
• 最新消息称，Meta正与AI初创公司Scale AI洽谈投资事宜，涉及金额有望超100亿美元，若达成将成为有史以来最大规模的私人公司融资之一。
• 此轮战略投资或标志着Meta进一步从元宇宙领域转向AI核心业务，提高其与谷歌、微软等科技巨头的竞争力。
• 截至目前，Meta尚未就相关谈判公开置评。

相关链接：

• Meta正就或超100亿美元的大规模AI投资谈判
• 消息称Meta正谈判投资Scale AI，预计价值超100亿美元

Meta Eyes Over $10 Billion AI Investment, Reportedly in Talks for Scale AI Stake 💰

Brief:

• According to informed sources, Meta is negotiating with potential partners for a large-scale AI investment exceeding $10 billion. The aim is to accelerate its research and development, and infrastructure building in areas such as generative AI and large language models. 💡
• Latest reports suggest Meta is discussing an investment with AI startup Scale AI, with the potential amount exceeding $10 billion. If finalized, it would become one of the largest private company financings ever. 📈
• This strategic investment could signal Meta's further pivot from the metaverse towards core AI businesses, enhancing its competitiveness against tech giants like Google and Microsoft.
• As of now, Meta has not publicly commented on the ongoing negotiations.

Related Links:

• Meta Negotiating Large-Scale AI Investment Potentially Exceeding $10 Billion
• Reports: Meta Negotiating Investment in Scale AI, Expected to Exceed $10 Billion

阿里发布千问3向量模型，性能超越国际顶尖AI模型

简报：

• 2024年6月6日，阿里巴巴正式开源千问3向量模型系列Qwen3-Embedding，针对文本表征、检索和排序任务进行优化。
• 新版模型在文本检索、聚类和分类等任务上较前代提升超40%，并在MTEB等专业榜单中超越谷歌Gemini Embedding、OpenAI text-embedding-3-large等顶级模型，获得同类最佳性能（SOTA）。
• 千问3向量模型支持超100种语言和多种编程语言，具备强大的多语言及跨语言能力，适用于全球开发者。
• 本次共开放9款不同尺寸及GGUF版本模型，并已在魔搭社区、Hugging Face与GitHub等平台上线，支持API调用与模块自定义。
• 自4月29日开源以来，千问3大模型已在多个国际榜单取得冠军成绩。

相关链接：

• 阿里开源千问3向量模型，性能大幅提升，挑战谷歌和OpenAI

Alibaba Releases Qwen3 Vector Model, Outperforming Top International AI Models

Briefing:

• On June 6, 2024, Alibaba officially open-sourced the Qwen3-Embedding vector model series, optimized for text representation, retrieval, and ranking tasks. 🚀
• The new model demonstrates over 40% improvement in tasks such as text retrieval, clustering, and classification compared to its predecessor. It has also surpassed top models like Google Gemini Embedding and OpenAI text-embedding-3-large on professional benchmarks like MTEB, achieving state-of-the-art (SOTA) performance in its category. 🏆
• Qwen3 vector models support over 100 languages and multiple programming languages, showcasing robust multilingual and cross-lingual capabilities, making them suitable for global developers.
• A total of 9 models of different sizes and GGUF versions have been released. They are now available on platforms such as ModelScope, Hugging Face, and GitHub, supporting API calls and module customization.
• Since its open-sourcing on April 29th, the Qwen3 large model has consistently achieved top rankings on various international benchmarks. 🌟

Related Links:

• Alibaba Open-Sources Qwen3 Vector Model, Significantly Boosting Performance and Challenging Google and OpenAI

台积电市值有望五年内突破3万亿美元，AI芯片需求成主要推动力

简报：

• 外媒The Motley Fool报道称，随着AI技术在智能手机、PC和数据中心等领域的广泛应用，以及台积电在晶圆代工2.0市场中的市占率上升，预计台积电市值有望在未来五年内达到3万亿美元。
• 台积电目前在全球第三方晶圆代工市场市占率高达67%，领先于三星等竞争对手。
• 台积电为英伟达、AMD、苹果等主要公司生产AI芯片，受益于全球AI芯片市场预计到2033年年复合增长率可达35%的趋势。
• 外媒预计，若台积电未来五年销售增长速度加快，市场或将给予其更高估值，使其市值突破3万亿美元大关。

相关链接：

• 外媒预测台积电市值有望五年内突破3万亿美元，AI芯片市场成关键推手

TSMC's Market Cap Poised to Exceed $3 Trillion Within Five Years, Driven by AI Chip Demand 🚀

Briefing:

• Foreign media outlet The Motley Fool reported that TSMC's market capitalization is expected to reach $3 trillion within the next five years. This projection is fueled by the widespread application of AI technology in areas such as smartphones, PCs, and data centers, as well as TSMC's increasing market share in the foundry 2.0 market.
• TSMC currently holds a dominant 67% market share in the global third-party wafer foundry market, leading competitors like Samsung.
• TSMC manufactures AI chips for major companies such as Nvidia, AMD, and Apple, benefiting significantly from the global AI chip market's projected compound annual growth rate (CAGR) of 35% until 2033. 📈
• Foreign media anticipates that if TSMC's sales growth accelerates over the next five years, the market might grant it a higher valuation, pushing its market cap past the $3 trillion mark. 💡

Related Link:

• Foreign Media Predicts TSMC Market Cap May Exceed $3 Trillion in Five Years, AI Chip Market Becomes Key Driver

国产具身大模型实现汽车制造多场景落地应用

简报：

• 2024年6月7日，智平方（深圳）科技有限公司与东风柳州汽车有限公司签署战略合作协议，双方将共同探索具身大模型在汽车制造各环节的首次深度应用。
• 智平方搭载全域全身视觉语言动作模型（VLA）的智能机器人AlphaBot 2已进驻东风柳汽工厂，在装配、物流、运维等全制造流程场景执行智能化作业，实现了国产具身大模型在汽车制造的全场景验证。
• 双方还将在智能驾驶与智能网联领域协同创新，推动智能驾驶技术产业化应用，提升整车智能化水平。

相关链接：

• IT之家报道
• 36氪快讯

Domestically Developed Embodied Large Models Achieve Multi-Scenario Application in Automotive Manufacturing 🚗

Briefing:

• On June 7, 2024, ZhiPingFang (Shenzhen) Technology Co., Ltd. and Dongfeng Liuzhou Motor Co., Ltd. signed a strategic cooperation agreement. Both parties will jointly explore the initial in-depth application of embodied large models across various stages of automotive manufacturing.
• ZhiPingFang's intelligent robot AlphaBot 2, equipped with a full-domain whole-body Vision-Language-Action (VLA) model, has been deployed at Dongfeng Liuzhou Motor's factory. It will perform intelligent operations across the entire manufacturing process, including assembly, logistics, and operations & maintenance (O&M), thereby achieving full-scenario validation of domestically developed embodied large models in automotive manufacturing. 🤖✨
• Both parties will also collaborate on innovation in the fields of intelligent driving and intelligent connected vehicles (ICV), promoting the industrial application of intelligent driving technology and enhancing the overall intelligence level of vehicles.

Related Links:

• IT Home Report (Chinese)
• 36Kr Newsflash (Chinese)

OpenAudio 推出轻量级开源多语言TTS模型S1-Mini

简报：

• Fish Audio 发布轻量级开源文本转语音模型 OpenAudio S1-Mini，仅0.5B参数，适用于资源受限设备，支持14种语言和50+种情感语调，能生成接近真人的超自然AI语音。
• S1-Mini 基于大规模数据集训练，经开放于 Hugging Face 平台，对开发者免费开放非商用下载和体验，降低了高品质语音合成门槛。
• 性能测试显示 S1-Mini 的自然度和情感表达在同类模型中表现突出，广泛应用于教育、娱乐、交互式应用等场景，预计将进一步推动开源TTS技术发展。

相关链接：

• 项目主页（Hugging Face）
• 新闻报道（AIbase）

OpenAudio Launches Lightweight Open-Source Multilingual TTS Model S1-Mini

Briefing:

• Fish Audio has released OpenAudio S1-Mini, a lightweight open-source text-to-speech (TTS) model. With only 0.5 billion parameters, it's ideal for resource-constrained devices. It supports 14 languages and over 50 emotional tones, capable of generating incredibly natural, human-like AI voices. 🗣️
• Trained on a massive dataset, S1-Mini is now available on the Hugging Face platform for free non-commercial download and use by developers, significantly lowering the barrier to high-quality speech synthesis. 🚀
• Performance tests show S1-Mini stands out among similar models for its naturalness and emotional expressiveness. It's expected to be widely applied in education, entertainment, interactive applications, and more, further boosting the development of open-source TTS technology. ✨

Related Links:

• Project Homepage (Hugging Face)
• News Report (AIbase)

OpenAI学院首个国际分院落地印度，推进AI教育普及

简报：

• OpenAI与印度政府电子与信息技术部下属IndiaAI本周宣布合作成立“OpenAI学院印度分院”，首次将OpenAI教育平台推广至国际市场。
• 合作聚焦提升印度全民AI技能，服务学生、开发者、教育者、官员、非营利机构和小企业主，并支持线上线下混合教学，初期提供英语与印地语，将扩展至更多地方语言。
• OpenAI将为印度FutureSkills平台与iGOT Karmayogi平台提供AI教育内容，在六大城市举办培训及工作坊，并向50家AI初创或研究项目每家提供最多10万美元API积分，并组织涉及7个邦、覆盖2.5万名学生的黑客马拉松。
• 计划为100万印度教师提供生成式AI工具培训，同时免费开放包括视频教程、实践模块在内的AI学习资源，促进AI知识在全球的公平获取。

相关链接：

• OpenAI与印度政府启动AI教育合作，OpenAI学院首次国际落地

OpenAI Academy's First International Branch Launches in India, Advancing AI Education Accessibility 🚀

Briefing:

• OpenAI, in collaboration with IndiaAI, a division under India's Ministry of Electronics and Information Technology, announced this week the establishment of the "OpenAI Academy India Branch." This marks the first time OpenAI's educational platform has expanded into an international market.
• The partnership aims to enhance AI skills for all Indians, serving students, developers, educators, government officials, non-profit organizations, and small business owners. It will support blended online and offline teaching, initially offering content in English and Hindi, with plans to expand to more local languages. 📚
• OpenAI will provide AI education content for India's FutureSkills platform and iGOT Karmayogi platform. Training and workshops will be held in six major cities, and up to $100,000 in API credits will be offered to each of 50 AI startups or research projects. Additionally, hackathons involving 7 states and covering 25,000 students will be organized.
• The initiative plans to train 1 million Indian teachers in generative AI tools. Concurrently, AI learning resources, including video tutorials and practical modules, will be made freely available, promoting equitable access to AI knowledge globally. ✨

Related Link:

• OpenAI and Indian Government Launch AI Education Partnership, OpenAI Academy Lands Internationally for the First Time