小米小爱换帅,MiMo终于接上超级小爱Xiaomi Xiaoai Gets a New Leader, MiMo Finally Catches Up to Super Xiaoai


AI free token/AI免费积分

本文来源:https://mp.weixin.qq.com/s/NFdBq6N7UGCAvih583Eoqg
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

小米把小爱同学又动了一刀。这次动的是技术归口,不是换了个产品经理那么简单。罗福莉领导的 MiMo 基座大模型团队,正式接管小爱同学的基础模型能力;干了近十年的原负责人王刚,调任机器人业务。我查了一下,这是小米把「明星语音助手」往「自研大模型主战场」并轨的最清晰一次组织调整。

图片

小爱同学对小米太重要了。2024 年升级成「超级小爱」,铺在手机、汽车、智能家居上,是「人车家全生态」的语音入口。雷军在发布会上说过,「在 AI 领域,我们力争成为国内乃至全球第一阵营的选手」。要把这句话落地,小爱背后的模型归谁管,比界面多几个按钮关键得多。

▎三条线拆清楚

调整后,小爱技术架构拆成三块。

第一块,基础模型,归罗福莉的 MiMo 团队。MiMo-V2-Pro、V2-Omni、V2-TTS 这一串自研模型,终于从「集团技术项目」变成「小爱发动机」。

第二块,云端工程化,由栾剑负责,包括 MiMo 在小爱侧的外部系统对接、云端能力搭建。栾剑早年就是小米大模型团队核心人物,这条线他接得住。

第三块,端侧能力,下放给手机、汽车等业务线的 OS 团队。意思是,小爱不再大包大揽所有终端适配,而是让离用户最近的系统团队自己把模型塞进场景里。

很多人不知道,王刚从小爱启动研发至今快十年,这次调任机器人,说明小米在 AI 人事上开始「专人专岗」,语音助手线和具身智能线分开抢时间。

图片

▎罗福莉接棒,不是挂名

罗福莉去年加入小米后,MiMo-V2-Flash、V2-Pro 接连上线,团队走极致扁平路线,全链路超 100 人,她甚至推行过「每天跟 Agent 对话少于 100 次可以辞职」这种狠规矩。她入职时在朋友圈写过一句话,「智能终将从语言迈向物理世界。我正在 Xiaomi MiMo,和一群富有创造力、才华横溢且真诚热爱的研究员,致力于构建这样的未来,全力奔赴我们心目中的 AGI。」

现在让她直接管小爱的基座模型,逻辑很顺。小爱缺的不是入口,是足够强的底层模型和 Agent 框架。今年 3 月上线的 miclaw,就是基于 MiMo 做的手机端智能体测试产品。卢伟冰说过,「miclaw 不会取代小爱同学,未来会跟超级小爱合体,界面还是超级小爱,能力会被 miclaw 赋能,变成更聪明、更能干的超级小爱。」

组织调整,就是把这句话从 PPT 写成汇报关系。模型团队、产品团队、终端团队各干各的,但发动机必须同一颗。

图片

▎跟 7 月中国区人事不是一回事

7 月 1 日小米中国区销售线刚换过人,王晓雁、陈幕楠那轮调整,打的是渠道和销量。这次是小爱 AI 技术线,两条线必须分开看。销售人事解决「货怎么卖」,罗福莉接棒解决「小爱够不够聪明」。

说实话,小米 AI 前几年存在感不如字节、阿里、华为扎眼。雷军今年口径是,2026 年 AI 研发和资本投入将超过 160 亿元人民币,未来三年还要砸 600 亿元。钱砸下去之后,组织怎么排兵,决定用户能不能在手机上感知到差距。

对小爱用户来说,短期界面可能变化不大,但后台如果真是 MiMo 统一供模,长文理解、跨 App 执行、车机手机联动的上限会抬高。小米接下来要证明的,不是又发了一个模型,而是超级小爱能不能在 daily 场景里打赢豆包、小艺、通义。

这轮换帅,本质是把小爱从「功能型助手」往「模型驱动型入口」推。罗福莉能不能把 MiMo 的纸面参数变成你每天喊一声就好用的体验,才是这刀组织调整的真正考题。

图片

 

Source: https://mp.weixin.qq.com/s/NFdBq6N7UGCAvih583Eoqg If there are any copyright issues, please contact us at 13951431913#139.com or submit a comment on the right.

Xiaomi has made another adjustment to Xiao Ai. This time, it's not a simple change of product manager, but a shift in technical responsibility. The MiMo foundational model team, led by Luo Fuli, has officially taken over the base model capabilities of Xiao Ai. The former head, Wang Gang, who has been working on it for nearly a decade, has been transferred to the robotics business. I checked, and this is the clearest move by Xiaomi to align the "star voice assistant" with the "in-house developed model battlefield".

图片

Xiao Ai is too important to Xiaomi. Upgraded to "Super Xiao Ai" in 2024, it has been deployed on phones, cars, and smart homes, serving as the voice interface for the "human-car-home full ecosystem." Lei Jun said at the launch event, "In the field of AI, we aim to be in the first tier domestically and globally." To make this statement a reality, who is in charge of the model behind Xiao Ai is much more critical than adding a few extra buttons to the interface.

▎Clearing up the three lines

After the adjustment, Xiao Ai's technical architecture is split into three parts.

The first is the base model, which belongs to Luo Fuli's MiMo team. The set of self-developed models like MiMo-V2-Pro, V2-Omni, and V2-TTS are finally transitioning from "corporate technology projects" to the "Xiao Ai engine."

The second is cloud-based engineering, led by Luan Jian, covering external system integration and cloud capability deployment for MiMo in the Xiao Ai context. Luan Jian was a core member of the Xiaomi model team early on, so he is well-equipped to handle this line.

The third is edge-side capability, decentralized to the OS teams of businesses like phones and cars. This means Xiao Ai will no longer take on all terminal adaptations itself; instead, the system teams closest to the users will embed the model into their respective scenarios.

Many people may not know that Wang Gang has been on the Xiao Ai project for nearly ten years. His transfer to robotics indicates that Xiaomi is beginning to specialize roles in AI, separating the voice assistant line from the embodied intelligence line to maximize efficiency.

图片

▎Luo Fuli steps up, not just a title

Since joining Xiaomi last year, Luo Fuli has seen the successful launch of MiMo-V2-Flash and V2-Pro. The team operates on an extremely flat structure with over 100 people in the full chain, and she even instituted a rule that employees must interact with an Agent over 100 times a day or face dismissal. In her welcome post on WeChat, she wrote, "Intelligence will inevitably move from language to the physical world. I am at Xiaomi MiMo, working with a group of creative, talented, and passionately committed researchers to build this future, fully pursuing the AGI we envision."

It makes perfect sense to have her directly manage Xiao Ai's base model now. What Xiao Ai lacks isn't entry points, but sufficiently strong underlying models and an Agent framework. In March this year, Xiaomi launched miclaw, a mobile-side smart agent test product based on MiMo. Lu Weibing stated, "miclaw will not replace Xiao Ai; in the future, it will merge with Super Xiao Ai, retaining the Super Xiao Ai interface while being empowered by miclaw to become a smarter and more capable Super Xiao Ai."

This organizational adjustment essentially turns that PPT statement into a hierarchy. The model team, product team, and terminal team each do their own thing, but they must all use the same engine.

图片

▎Unrelated to the July China region personnel changes

The personnel changes in the Xiaomi China region sales line (Wang Xiaoyan, Chen Munan) that occurred on July 1 were focused on channels and sales volume. This adjustment pertains to the AI technical line of Xiao Ai; the two lines must be viewed separately. The sales adjustment addressed "how to sell the product," while Luo Fuli's appointment addresses "how intelligent Xiao Ai is."

To be honest, Xiaomi's AI presence in recent years hasn't been as strong as that of ByteDance, Alibaba, or Huawei. Lei Jun recently announced that AI research and capital investment will exceed 16 billion RMB in 2026, with a total of 60 billion RMB planned for the next three years. Once the money is spent, the organizational structure determines whether users will notice a difference in their phone experiences.

For Xiao Ai users, the interface might not change much in the short term, but if the backend truly unifies under the MiMo supply, the limits of long-text understanding, cross-app execution, and car-mobile interconnection will be raised. What Xiaomi needs to prove isn't that they released another model, but that Super Xiao Ai can outperform Doubao, Xiaoyi, and Tongyi in daily-life scenarios.

This reshuffle is fundamentally about pushing Xiao Ai from a "functional assistant" to a "model-driven entry point." Whether Luo Fuli can translate MiMo's paper specifications into a usable experience with a single voice command is the real test of this organizational change.

图片

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry