![[a8f36b5b-7a9d-4a40-9779-92c28d6af8dc.mp3]]

松下发布OmniFlow，实现多模态AI低成本任意数据转换

简报：

• 松下控股公司联合美国松下研发公司及UCLA研究团队开发出“OmniFlow”多模态生成AI，支持文本、图像与音频之间的任意转换，大幅提升多模态AI的灵活性和应用潜力。
• OmniFlow可在训练数据极少的情况下实现高精度转换，降低数据采集成本至传统方案的1/60。
• 该技术将在2025年CVPR会议上发布，被业界认为有望广泛应用于工厂自动化和生活服务等多个领域。
• OmniFlow通过三种数据特征的深度连接与处理，提升了多模态生成任务的表现，优于当前主流方法。

相关链接：

• 松下推出OmniFlow多模态生成AI，实现“文本-图像-音频”任意转换

Panasonic Unveils OmniFlow: Low-Cost, Arbitrary Data Conversion for Multimodal AI

Brief:

• Panasonic Holdings, in collaboration with Panasonic R&D Company of America and a UCLA research team, has developed "OmniFlow," a multimodal generative AI. It enables arbitrary conversion between text, images, and audio, significantly enhancing the flexibility and application potential of multimodal AI. ✨
• OmniFlow achieves high-precision conversions with minimal training data, reducing data acquisition costs to 1/60th of traditional methods. 💰
• The technology will be presented at the 2025 CVPR conference and is expected by the industry to find wide applications in various fields, including factory automation and daily life services. 🤖
• OmniFlow improves the performance of multimodal generative tasks by deeply connecting and processing three types of data features, outperforming current mainstream methods.

Related Links:

• Panasonic launches OmniFlow multimodal generative AI, enabling arbitrary conversion of “text-image-audio”

OpenAI收购Windsurf后开发者因模型访问受限表达不满

简报：

• OpenAI以约30亿美元收购了AI编码初创公司Windsurf，收购后Windsurf用户发现对Anthropic的Claude3.7Sonnet和Claude3.5Sonnet模型的访问权限大幅缩减。
• 这一变化导致开发者面临工具受限、使用成本上升和操作流程复杂等问题，用户对未来增长潜力产生担忧。
• Windsurf目前未获得Claude4系列模型的直接访问权限，影响平台的竞争力和用户体验，引发开发者对OpenAI收购后便利性的质疑。
• 若不能恢复对模型的访问和提升服务质量，Windsurf或将面临用户流失与市场劣势的风险。

相关链接：

• OpenAI收购Windsurf引发开发者不满——aibase新闻
• OpenAI收购Windsurf后部分AI模型访问权限收紧
• OpenAI收购Windsurf引发开发者争议：Claude模型受限
• OpenAI以30亿美元收购Windsurf，用户呼吁恢复模型支持
• OpenAI以30亿美元收购AI编程工具Windsurf，使用者称模型访问受限影响体验

Developers Express Discontent Over Restricted Model Access Following OpenAI's Acquisition of Windsurf

Brief:

• OpenAI has acquired AI coding startup Windsurf for approximately $3 billion. 💰 Following the acquisition, Windsurf users have reported a significant reduction in access to Anthropic's Claude 3.7 Sonnet and Claude 3.5 Sonnet models.
• This change has led developers to face issues such as limited tools, increased usage costs, and complex operational procedures, raising concerns about future growth potential among users. 😟
• Windsurf currently does not have direct access to the Claude 4 series models, impacting the platform's competitiveness and user experience. This has prompted developers to question the convenience of the platform after OpenAI's acquisition.
• If access to these models is not restored and service quality is not improved, Windsurf may face the risk of user attrition and market disadvantage. 📉

Related Links:

• OpenAI收购Windsurf引发开发者不满——aibase新闻
• OpenAI收购Windsurf后部分AI模型访问权限收紧
• OpenAI收购Windsurf引发开发者争议：Claude模型受限
• OpenAI以30亿美元收购Windsurf，用户呼吁恢复模型支持
• OpenAI以30亿美元收购AI编程工具Windsurf，使用者称模型访问受限影响体验

谷歌推AI Edge Gallery，手机本地可离线运行多种AI模型

简报：

• 谷歌低调发布实验性应用Google AI Edge Gallery，使用户可在Android设备上无需网络离线运行开源AI模型，实现图像生成、文本处理和代码编辑等多种功能。
• 应用集成Hugging Face平台，支持从500MB到4GB等多规格模型，全部在本地运行，强化隐私保护和响应速度。
• 当前仅支持Android，iOS版本即将推出，用户可通过GitHub下载APK安装。
• 性能测试显示，其离线表现可媲美早期主流云端AI模型。

相关链接：

• AI Edge Gallery应用介绍-爱必思
• AI Edge Gallery功能解析-IT之家
• 谷歌AI Edge Gallery支持说明

Google Launches AI Edge Gallery, Enabling Offline AI Model Execution on Mobile Devices

Briefing:

• Google has quietly released an experimental app, Google AI Edge Gallery, allowing users to run open-source AI models offline on Android devices without an internet connection, enabling various functions such as image generation, text processing, and code editing. 💡
• The app integrates with the Hugging Face platform, supporting models ranging from 500MB to 4GB, all running locally, enhancing privacy protection and response speed. ⚡
• Currently, it only supports Android, with an iOS version coming soon. Users can download the APK for installation via GitHub.
• Performance tests indicate that its offline capabilities are comparable to early mainstream cloud-based AI models. 🚀

Related Links:

• Introduction to AI Edge Gallery App - Aibase
• AI Edge Gallery Feature Analysis - IT Home
• Google AI Edge Gallery Support Information

台积电1.4纳米制程晶圆价格创历史新高，达4.5万美元

简报：

• 台积电即将推出的1.4纳米Angstrom制程每片晶圆成本高达4.5万美元（约新台币145万元），比现有2纳米制程贵出约5成，刷新历史最高纪录。
• 即便苹果、联发科、高通等长期合作的VIP客户，也对如此高昂的价格感到犹豫，目前尚无客户正式表态投片1.4纳米。
• 先进制程制造成本激增的原因包括技术复杂度提升和ASML新一代极紫外光（EUV）设备成本高昂。
• 目前2纳米制程晶圆报价为3万美元，自2024年4月起已开放接单，联发科和苹果抢先卡位。
• 第三代3纳米N3E制程也被多家顶级芯片公司采用，成为当前市场关注焦点。

相关链接：

• 台积电1.4奈米成本爆4.5万美元创新高
• 台积电2奈米制程最新进展
• 3纳米制程市场动向

TSMC's 1.4nm Wafer Price Hits Record High at $45,000

Briefing:

• TSMC's upcoming 1.4nm Angstrom process wafers are priced at an astonishing $45,000 (approximately NT$1.45 million) each, a roughly 50% increase compared to the existing 2nm process, setting a new historical record. 🤯
• Even long-term VIP clients like Apple, MediaTek, and Qualcomm are hesitant about such exorbitant prices, with no customers officially committing to 1.4nm tape-outs yet.
• The surging manufacturing costs for advanced processes are attributed to increased technical complexity and the high price of ASML's next-generation Extreme Ultraviolet (EUV) equipment.
• Currently, 2nm process wafers are quoted at $30,000 and have been open for orders since April 2024, with MediaTek and Apple leading the charge. 🚀
• The third-generation 3nm N3E process is also being adopted by several top-tier chip companies, becoming a current market focal point. ✨

Related Links:

• 台积电1.4奈米成本爆4.5万美元创新高
• 台积电2奈米制程最新进展
• 3纳米制程市场动向

本吉奥成立LawZero，开发“科学家AI”防止智能体欺骗

简报：

• AI领域知名学者约书亚·本吉奥近日宣布成立非营利组织LawZero，启动资金约3000万美元，致力于开发一套被称为“科学家AI”的系统，用于防范AI智能体欺骗用户。
• “科学家AI”系统定位为AI安全的护栏，其智能水平将至少与所监管的AI持平，以便于识别并制止潜在的欺骗行为，并保障AI运行过程的透明性和可信度。
• 本吉奥表示，LawZero希望通过该系统为AI应用建立更高的道德标准，并提升行业对AI安全和责任的关注。

相关链接：

• aibase.com新闻报道1

Yoshua Bengio Founds LawZero to Develop "Scientist AI" to Prevent Agent Deception

Brief:

• Renowned AI scholar Yoshua Bengio recently announced the establishment of LawZero, a non-profit organization with initial funding of approximately $30 million. Its mission is to develop a system dubbed "Scientist AI" aimed at preventing AI agents from deceiving users. 🤖
• The "Scientist AI" system is positioned as a guardrail for AI safety. Its intelligence level will be at least on par with the AI it supervises, enabling it to identify and halt potential deceptive behaviors, while also ensuring transparency and trustworthiness in AI operations. ✨
• Bengio stated that LawZero hopes to establish higher ethical standards for AI applications through this system and elevate the industry's focus on AI safety and responsibility. 🛡️

Related Links:

• aibase.com News Report 1