![[7674007a-598c-4b44-a93b-95be2dc6bb28.mp3]]

英伟达以4亿美元收购CentML，95后华人团队加入加速AI优化

简报：

• 英伟达以4亿美元收购位于多伦多的AI初创公司CentML，该公司由95后华人王尚及其导师于2023年创立，专注于AI模型优化与GPU性能提升。
• 此次收购中，王尚及其团队（包括3位联合创始人和15名工程师）整体加入英伟达，王尚将负责AI系统软件管理，其团队被认为为英伟达CUDA工具链注入新活力。
• CentML此前已获英伟达和谷歌风险基金2700万美元投资，因其帮助用户更高效利用硬件资源，在AI模型优化领域迅速崭露头角。
• 英伟达通过密集收购创业公司来聚集顶尖AI人才，进一步巩固其在全球AI领域的领先地位。

相关链接：

• aibase.com新闻报道1

Nvidia Acquires CentML for $400 Million, Young Chinese Team Joins to Accelerate AI Optimization 🚀

Brief:

• Nvidia has acquired CentML, a Toronto-based AI startup, for $400 million. Founded in 2023 by Chinese national Wang Shang (born post-1995) and his mentor, CentML specializes in AI model optimization and enhancing GPU performance.
• As part of the acquisition, Wang Shang and his entire team (including 3 co-founders and 15 engineers) will join Nvidia. Wang Shang will lead AI system software management, and his team is expected to inject new vitality into Nvidia's CUDA toolchain.
• CentML had previously secured $27 million in investments from Nvidia and Google venture funds. It rapidly gained prominence in the AI model optimization field by enabling users to utilize hardware resources more efficiently. 💡
• Nvidia continues to solidify its leading position in the global AI sector by actively acquiring startups to gather top AI talent. 💰

Related Links:

• aibase.com News Report 1

Meta测试可主动联系用户的AI聊天机器人，支持定制与个性化跟进

简报：

• Meta正在测试一类新型聊天机器人，能主动向用户发送消息，而不仅仅是被动回复，已在Facebook Messenger等平台尝试应用。
• 用户可在Meta AI Studio平台定制个人专属AI机器人，机器人能够根据用户偏好，14天内自动发起基于先前对话内容的联系。
• 机器人可设置为私有，也能通过社交渠道展示或分享，被设计用于辅助用户发现兴趣内容、增强互动体验。
• Meta表示这些AI不具备专业咨询能力，风险防控措施尚不明确，同时公司计划借此创新AI服务在未来实现20亿至30亿美元收入目标。
• 相关安全和隐私问题引发行业关注，目前尚待进一步观察。

相关链接：

• Meta测试主动式AI聊天机器人，支持定制化跟进
• IT之家报道：Meta AI聊天机器人将主动联系用户

Meta Tests AI Chatbots That Can Proactively Contact Users, Supporting Customization and Personalized Follow-up

Briefing:

• Meta is testing a new type of chatbot that can proactively send messages to users, rather than just passively responding. This feature is already being trialed on platforms like Facebook Messenger. 🤖
• Users can customize their own AI chatbots on the Meta AI Studio platform. These bots can automatically initiate contact based on previous conversations within 14 days, according to user preferences.
• The chatbots can be set to private or displayed/shared via social channels. They are designed to help users discover content of interest and enhance interactive experiences. ✨
• Meta states that these AIs do not possess professional advisory capabilities, and risk control measures are not yet clear. Simultaneously, the company plans to leverage these innovative AI services to achieve a future revenue target of $2-3 billion.
• Related safety and privacy concerns have drawn industry attention and are currently awaiting further observation. 🔒

Related Links:

• Meta tests proactive AI chatbots, supporting customized follow-up
• IT Home Report: Meta AI Chatbots Will Proactively Contact Users

欧盟AI法案按时实施，科技巨头合规压力骤增

简报：

• 欧盟确认其划时代的《人工智能法案》将于2026年年中正式生效，拒绝任何“宽限期”或暂停要求，驳回了包括Alphabet、Meta等逾百家科技公司的推迟游说。
• 此法案采取基于风险的监管模式，明确定义“不可接受风险”和“高风险”AI应用范畴，对生物识别、教育、就业等高风险领域实施注册和风险管理要求，准入欧盟市场前必须合规。
• 近年来欧盟已分阶段推动相关立法，全面规则落地后，全球科技企业必须在合规压力下加快调整，确保其AI产品符合欧盟标准。

相关链接：

• AIbase新闻详情1
• AIbase新闻详情2
• AIbase新闻详情3

EU AI Act on Schedule, Piling Pressure on Tech Giants ⚖️🚀

Briefing:

• The EU has confirmed its landmark Artificial Intelligence Act will officially come into force by mid-2026, rejecting calls for any "grace periods" or postponements. This dismisses lobbying efforts from over a hundred tech companies, including Alphabet and Meta, to delay implementation. 🇪🇺
• The Act adopts a risk-based regulatory model, clearly defining "unacceptable risk" and "high-risk" AI application categories. It imposes strict registration and risk management requirements for high-risk areas such as biometrics, education, and employment. Compliance is mandatory before market entry in the EU.
• In recent years, the EU has progressively advanced relevant legislation in phases. With the full implementation of these comprehensive rules, global tech companies must accelerate adjustments under increasing compliance pressure to ensure their AI products meet EU standards.

Related Links:

• AIbase News Detail 1
• AIbase News Detail 2
• AIbase News Detail 3

B站开源AniSora V3升级：动漫视频生成质量与效率双提升

简报：

• B站近日推出动漫视频生成模型AniSora V3，针对动漫、漫画和VTuber内容创作，显著提升了视频质量、动作流畅度和风格多样性。
• AniSora V3基于CogVideoX-5B和Wan2.1-14B模型，结合强化学习与人类反馈（RLHF）框架，实现了业内优秀的角色一致性与动作自然度。
• 新版支持华为Ascend910B NPU，推理速度提升约20%，生成4秒视频仅需2-3分钟，多任务能力增强，支持单帧转视频、关键帧插值、唇部同步等应用。
• AniSora V3可生成高达1080p分辨率的视频，涵盖日本动漫、国创动画、漫画改编等多种风格，适用于预告片、漫画动画化和游戏角色动作等场景。

相关链接：

• IT之家新闻原文
• AniSora V3开源项目地址

Bilibili Open-sources AniSora V3 Upgrade: Dual Improvement in Anime Video Generation Quality and Efficiency 🚀

Briefing:

• Bilibili recently launched its anime video generation model, AniSora V3, specifically designed for anime, manga, and VTuber content creation. This new version significantly enhances video quality, motion fluidity, and style diversity.
• AniSora V3 is built upon the CogVideoX-5B and Wan2.1-14B models, integrating a Reinforcement Learning from Human Feedback (RLHF) framework to achieve industry-leading character consistency and natural motion. ✨
• The new version supports the Huawei Ascend 910B NPU, boosting inference speed by approximately 20%. Generating a 4-second video now takes only 2-3 minutes. Its multi-task capabilities have also been enhanced, supporting applications like single-frame-to-video conversion, keyframe interpolation, and lip synchronization.
• AniSora V3 can generate videos up to 1080p resolution, covering various styles including Japanese anime, Chinese original animation, and manga adaptations. It is suitable for scenarios such as trailer creation, manga animation, and game character motion. 🎬

Related Links:

• IT Home News Article
• AniSora V3 Open-source Project Address

B站发布“代号H”AI工具，助力播客视频化并推三大扶持政策

简报：

• B站近日宣布将大力发展视频播客业务，并推出名为“代号H”的AI创作工具，支持文案和音频输入，帮助创作者高效生成视频内容。
• 官方将同步推出流量冷启动支持、免费一线城市录制场地等扶持举措，推动音频及文字作者向视频化转型。
• “代号H”AI工具目前支持千字内容6分钟内自动生成视频，创作者体验反馈“超出预期”，预计未来可进一步提速至3分钟内完成。
• 2025年第一季度B站视频播客消费时长超259亿分钟，同比增长270%，用户规模突破4000万。

相关链接：

• B站推出“代号H”AI创作工具，全面发力视频播客
• 消息称B站推出AI创作工具“代号H”

Bilibili Launches "Project H" AI Tool to Empower Podcast Videoization and Introduce Three Major Support Policies

Brief:

• Bilibili recently announced its strong commitment to developing its video podcast business, launching an AI creation tool called "Project H" that supports text and audio input to help creators efficiently generate video content. 🚀
• Concurrently, Bilibili will introduce supportive measures such as traffic cold-start assistance and free recording studios in first-tier cities to facilitate the transition of audio and text creators towards video formats.
• The "Project H" AI tool currently supports automatic video generation from 1,000 characters of content within 6 minutes. Creator feedback has been "beyond expectation," and it's anticipated that future versions will further accelerate this to under 3 minutes. ⚡️
• In Q1 2025, Bilibili's video podcast consumption time exceeded 25.9 billion minutes, marking a 270% year-on-year growth, with the user base surpassing 40 million. 📈

Related Links:

• Bilibili launches "Project H" AI creation tool, fully expanding into video podcasts
• Report: Bilibili launches AI creation tool "Project H"