Tiktok can't catch up with Sora

Alphabet List

follow

April 22, 2024

0 comments

738 browsing

4 Collections

19 minutes

  B-end product managers need to conduct more in-depth user interviews, research, and analysis, while C-end product managers need more rapid user testing, feedback, and iteration

Back to the big tree of Tiktok, Clipping had a great opportunity to take the lead in the field of AI Wensheng video. But after OpenAI released Sora, I was caught off guard; Even if it's striving to catch up, OpenAI still takes the lead, and even the entire ByteDance has slowed down.

On the popular track of AI Wensheng Video, the clip of Tiktok is being thrown away by Sora of OpenAI.

Recently, American software giant Adobe announced that it will add multiple cultural and creative video AI tools to the new version of the well-known video editing software Premiere Pro. Sora, which has shaken the global technology industry in the past two months, as well as two similar products Gen-2 and Pika, will soon join the Adobe Family Bucket.

With the help of third-party AI tools such as Sora, the new version of Premiere Pro can not only edit and process pre shot regular videos, but also generate AI videos in real-time based on user input text, and integrate the two into one.

Adobe has released an official demonstration video: a man walks up to the window and watches the night view of the entire city. Users do not need to shoot real-life scenes, they only need to input a piece of text to generate a city rain night video using Sora, which seamlessly connects with the previous video, and the effect can be almost lifelike.

Sora made its debut in February this year, and OpenAI released several demonstration videos, but did not disclose the product progress and launch time. Nowadays, Sora has been accepted by Adobe, indicating significant progress in the past few months and being closer to open use.

On the other hand, the video editing software behind Tiktok is also moving towards AIGC (Artificial Intelligence Generated Content). But as of now, Jianying has not yet produced stunning results.

At present, there are many AI gameplay options for Jianying, including one click filming, cutting the same style, AI cloning sound, digital population broadcasting, etc. However, it does not have the ability to directly generate videos based on text. Its overseas version, Capcut, launched its Bunsen video feature at the end of February, but the effect is far from Sora.

Nowadays, Sora is not only leading in technology and products, but also favored by Adobe. Adobe has over 33 million paying users worldwide; Sora's integration into Adobe Family Bucket is expected to generate subscription revenue sharing and thus establish a preliminary business model.

This also means that for Jian Ying, who is still polishing AIGC capabilities, catching up with Sora will become increasingly difficult.

Why did I transfer to a product manager after working in interaction design for 10 years?

After the real job transfer, I found that many jobs were still beyond my imagination. The job of a product manager is indeed quite complicated. In theory, the job of a product manager includes all aspects of the product, including market research, user research, data analysis

View details>

Jianying is regarded by the outside world as the key chip for Tiktok to catch the AIGC era express. Thanks to the blessing of Tiktok, Clipping has become one of the mobile video editing software with the largest number of users, producing a large number of short videos every day; Building upon this foundation, it seems natural to move from UGC (User Generated Content) to AIGC.

On February 7 this year, Zhang Nan, a veteran of Tiktok's meritorious service, resigned as CEO of the Group and led the team to cut the film. In an internal letter announcing her job change, she stated that AI image generation had a great impact on her and had enormous potential. She decided to "let go of everything" and set out without hesitation.

The development foundation of Jianying is not bad, and Tiktok is also highly valued. But more than two months have passed, and Jianying still hasn't made much progress, while Sora continues to make great strides.

In the AIGC era, Tiktok and the bytes behind it always seem to play slowly.

Eight years ago, ByteDance laid out its AI and established a specialized laboratory to recruit a large number of industry elites, but its achievements remained in content review, automatic translation, search services, and so on. In the second half of 2022, ChatGPT will ignite the industry; ByteDance immediately increased its investment in large models and launched over a dozen applications, including AI chatbots, but has yet to come up with the technology and products that will shake the industry.

At the annual staff meeting at the end of January this year, ByteDance CEO Liang Rubo exclaimed that "all the big companies that ByteDance should have had are sick.". He specifically named the AI business and said, "The semi annual technical review at the company level did not start discussing GPT until 2023, and the well done large model startups in the industry were all founded from 2018 to 2021."

Nowadays, in the field of video content, OpenAI is leading the way, and even the entire ByteDance industry is slowing down.

01

Back to the big tree of Tiktok, Clipping had a great opportunity to take the lead in the field of AI Wensheng video.

Jianying was launched in 2019 and a professional version for PC in 2021. It significantly lowers the threshold of UGC video creation. Ordinary people can generate a video of acceptable quality and release it to Tiktok with one click just by preparing the material and clicking a few times.

In addition, as a tool software, Jianying also has certain community attributes. In addition to the official video creation classroom, users can also refer to the video templates created by experts for creation, that is, "cut and match". This helps to improve user retention and activity.

Relying on its rich functions and deep binding with Tiktok, Jianying has won more than 100 million active users in just three years, jumping to the first place in similar software.

The growth of the overseas version of CapCut is also astonishing.

Capcut was launched in 2020 and is closely related to TikTok. According to data from the mobile application analysis platform, Capcut currently has over 200 million monthly active users. According to market research firm data.ai, as of August 2023, Capcut has over 490 million iPhone and Android users, equivalent to 1/4 of TikTok's global user base.

In addition to its large user base, another advantage of Jianying in the field of AI is that it has a large amount of video data that can be used for training large AI models.

Data, algorithms, and computing power are the three fundamental elements of the AI big model, with data being the foundation. In order to improve the comprehensive ability of the big model, we need to constantly "feed" multimodal data, mainly Internet text, images and videos.

Companies such as OpenAI and Adobe do not directly control data and must obtain it through third-party payment, which is costly. According to media reports, OpenAI spends as much as $1-5 million annually on purchasing copyrighted news article authorizations alone; Adobe purchased video clips from another large model service, Midjournal, at a price of $3 per minute to train its own cultural and educational video model.

In contrast, Scissor Image is tied to Tiktok in China and backed by TikTok overseas. Every day, a large number of users use it to edit and upload videos. This allows Jianying to access a large amount of video content at a relatively low cost, laying the foundation for training large models and exploring AI cultural and biological video functions.

However, since 2019, Jianying has not developed a strong AIGC capability, and has instead started commercialization early.

Jianying has been a VIP member for a long time, and users pay 20-30 yuan per month to use exclusive materials, selected templates, etc. AI gameplay is also listed as one of the member benefits. According to the official introduction, AI gameplay mainly includes "infinite creation", "infinite mirror movement" and "instant universe", and AI beautification editing based on existing video materials.

It is not difficult to see that the AI gameplay of Jianying is far from that of literary videos like Sora. In addition, its fees are not cheap: members can earn 1200 points per month, while using the Infinite Mirror requires 480 points. After the points are depleted, users can choose to continue recharging, with a redemption ratio of 1 yuan to 100 points.

At the beginning of the birth of Scissor Film, the goal was to lower the threshold of short video creation as much as possible and promote the prosperity of Tiktok UGC ecology. In the past five years, it has indeed completed this task. Behind a large number of videos of sound shaking blockbusters, there are cutting and screening technologies and templates. Especially for those interesting special effect videos and stampede videos, after the release of Tiktok celebrities, a large number of users quickly followed suit and jointly promoted the popularity; Without the help of Jianying, it may not be easy to achieve this.

But now it seems that Tiktok still sets the ceiling of the clip too low. It had the opportunity to become a groundbreaking product like Sora, but it has always remained in the realm of video editing software.

Since the beginning of this year, Jian Ying has been striving to catch up, but the opportunity has been lost, the opponent has arrived, and the difficulty of catching up has rapidly increased. Jianying was left behind by Sora, and ByteDance still hasn't been able to break free from the cycle of "getting up early and rushing to the late episodes" in the AI field. This also indirectly confirms Liang Rubo's previous criticism of "mediocre gravity".

02

Overemphasizing the importance of helping the business and pursuing commercialization too early is a challenge for Jianying and even the entire ByteDance AI sector.

In addition to offering paid members and incorporating AI gameplay into the benefits package, Jianying has also added many advertising spaces within the app. For example, when users click "cut the same style", in addition to mobile phone photos and video materials, there are also banner advertisements floating above the materials, which are almost unrelated.

As a tool software with over 100 million users, the regular investment and financial pressure of Jianying will not be particularly significant. The reason why it developed membership and advertising early may be related to ByteDance's work style and evaluation criteria.

As we all know, byte is a super fast paced and highly involved Internet giant. If individuals, teams, or businesses cannot quickly bring visible output, they may be adjusted or even completely eliminated. Even AI that requires long-term investment cannot break away from this invisible benchmark of evaluation.

As early as 2016, ByteDance established an AI Lab artificial intelligence laboratory and introduced multiple academic and industry elites. At that time, OpenAI was also emerging and moving towards the vision of universal artificial intelligence, viewing itself as a non-profit organization.

In contrast, although AI Lab is called a "laboratory", it still requires close cooperation and service to the business. Its official website claims that its research focus is on developing new technologies for serving ByteDance content platforms; The specific fields include natural language processing, data mining, computer vision, machine learning, etc., and they are closely connected with the business segments of Tiktok.

In the following years, Byte came up with a series of AI tools, such as Byte Translator, an AI manuscript writing robot Xiaomingbot, and search services for Toutiao and Tiktok. They are certainly valuable, but they cannot be considered innovative products that break through the cognitive boundaries of the AI field and define the paradigm of AI development.

Until the second half of 2022, OpenAI sharpened its sword for many years, ChatGPT swept the world, and generative AI became the focus of competition for global technology companies. It was not until then that ByteDance devoted more energy to this new wave.

Zhang Yiming, who has faded out of ByteDance's frontline management, has a strong interest in AI and encourages the team to invest heavily. ByteDance cannot miss AGI (General Artificial Intelligence), which is an indispensable partner for Tiktok and TikTok to discover new growth opportunities in the world.

After the founder spoke, various business departments of ByteDance were moved by the news. Since then, ByteDance has launched more than ten AI products, such as Doubao, Tonglu, Kouzi, Gauth, etc., and AI features have also been added to Jianying, Feishu, and others.

But in this round of massive investment, ByteDance's logic of starting from business needs and benchmarking against competitors has been continued. Doubao and other products are scenario based applications of existing AI technology, rather than native exploration of AGI.

For example, in March last year, Microsoft launched the 365 Copilot, which integrated GPT functionality, causing a shock in the global office software market. A month later, Feishu announced the upcoming launch of its AI assistant "My AI", which provides various functions in the form of dialogue, including optimizing and continuing text content, creating schedules, automatically summarizing meeting minutes, and searching the company's internal knowledge base.

A year later, GPT has already blossomed everywhere in Microsoft's "family bucket", driving its stock price from $250 to over $400. However, Feishu was unable to rely on My AI to complete its comeback and instead announced layoffs at the end of March.

For example, Sora released a demonstration video on February 16th this year, and CapCut announced the launch of this feature a week later. Each person can generate 5 free videos per day. CapCut's technological prowess is clearly incomparable to OpenAI, and its cultural and video functions are also relatively rudimentary; Rushing to launch and benchmarking Sora inevitably carries the smell of seizing hot topics and forcibly completing KPIs.

Since Zhang Yiming named AGI, ByteAI has gone through another year without significantly narrowing the gap with OpenAI, and even showing a trend of expansion. The excessive tilt of ByteAI towards pragmatism not only misses out on past opportunities, but may also drag down the pace of catching up.

03

ByteDance has already noticed the situation and harm of the AI sector fighting on its own and revolving around its business.

In November 2023, ByteDance drew elite personnel from multiple departments to form the AI department Flow. TikTok's technical leader Zhu Wenjia, Byte's Vice President of Products and Strategy Zhu Jun, Byte's Vice President of Technology Hong Dingkun, and Feishu's Vice President of Products Qi Junyuan all joined in. The high configuration of personnel reveals ByteDance's intention to coordinate AI development and eliminate duplicate construction through Flow.

At present, the highest exposure AI products of ByteDance, such as Doubao, Button, and Phone Stove, are all managed by the Flow department. Although there are AI businesses such as Jianying, Feishu, and Dali Education, their popularity has gradually been overshadowed by Flow.

On the other hand, the Byte Skylark model has been polished for nearly a year, and the multi-modal model BuboGPT has also made progress, laying the foundation for the mass production of AI applications in Flow. Over time, Flow is expected to reverse the situation where ByteAI is always one step slower.

However, ByteDance is not OpenAI after all, and there are no super capitalists like Microsoft. Although AGI is important, its impact on existing businesses is not yet significant and requires longer sedimentation and polishing to unleash commercial value; Byte's investment in AGI cannot be endless, and the input-output ratio in the medium and short term must be considered.

In the short term, there are many places where bytes need money, people and resources, such as Tiktok shelf e-commerce, life services, etc., which need a lot of real money and silver. The following phenomenon is that although Flow has taken the lead, ByteAI still shows a tendency to balance business needs.

According to the recent report of Tech Planet, Tiktok Life Service has just set up an AI team, hoping to create incremental business value by using AI technology, and has started the research and development of relevant AI products, including building a content creation platform related to life services.

The reason why Tiktok Life Service has pulled up such a team may be related to the action of competitors Meituan and Hunglema.

Meituan has recently started small-scale testing of its AI assistant service "Ask Little Bag", which can recommend takeout products that meet users' needs and provide dining suggestions. Ele.me released its "AI Business Assistant" to retail industry merchants in early April, which can intelligently generate various key business reports and data for merchants.

In this case, it may be one of the basic goals of Tiktok Life Service's entry into AI to stop Meituan from being too hungry to focus on it.

Under the influence of internal and external factors, ByteDance has no conditions to accumulate and develop like OpenAI. The dual goal of ByteAI is to catch up with the AGI trend while also having the ability to quickly implement and serve the business.

Under the supervision of Zhang Yiming and Liang Rubo, there is no need for outsiders to question ByteDance's determination to do AI. But if in another year or two, ByteDance still cannot keep up with the pace and level of OpenAI, then it may consider another option: to return to the role of a "water seller" and become an AGI training material provider.

As mentioned earlier, data is one of the three basic elements of the big model, and today's headlines, Tiktok, TikTok, etc. under Byte have accumulated hundreds of millions of texts, pictures and videos. ByteDance can use this data to train its own large models, or consider further selling it to third-party companies such as OpenAI, while properly addressing security and privacy issues.

Acting as a "water seller" in the AGI era is actually a variant of ByteDance's expertise in traffic business. ByteDance's current cash cow - advertising and e-commerce businesses - are built on the basis of monetization of traffic; If the AI company is transformed into a new customer, then ByteDance's second growth curve problem will be easily solved.

On the other hand, since OpenAI can collaborate with Microsoft and Adobe, it is not unimaginable to collaborate with ByteDance. After all, Tiktok and TikTok are the largest landing scenarios for Sora and other AIGC services. If Byte and OpenAI reach a "competitive" relationship, Byte will break out of a slow and slow cycle and catch the AI fast train with the help of OpenAI.

Reference materials:

Alphabet List, ByteAI: Lost PICO, Received Feishu

Alphabet chart, "Will Sora" kill "Jianying?"

Light cone intelligence, "Not willing to just be an AI application factory, ByteDance make-up model"

Tech Planet, Meituan Tiktok Bets on a New Battlefield: Inside Test and Take out Assistant "Ask a Small Bag", Set up Life Service AI

Xin Zhiyuan, "33 million editors are revolutionized, Sora, Pika, and Gen-2 will be fully launched on Adobe"

Author: Yan Fei, Editor: Wang Jing

Source official account: Alphabet list (ID: wujicaijing), making the future more than big.

This article is authorized for publication by the collaborative media @ Alphabet, in which everyone is a product manager. Reproduction without permission is prohibited.

The question is from Unsplash and is based on the CC0 protocol

The viewpoint of this article only represents the author himself, and everyone is a product manager. The platform only provides information storage space services.

抖音追不上Sora

字母榜关注

2024-04-22

0 评论738 浏览4 收藏19 分钟

背靠抖音这颗大树，剪映原本有很大机会在AI文生视频领域的占得先机。但OpenAI发布Sora之后，被打了个措手不及；即便是奋起追赶，OpenAI还是抢先，剪映乃至整个字节又双叒慢了。

在AI文生视频这条热门赛道，抖音旗下的剪映，正在被OpenAI的Sora越甩越远。

近日，美国软件巨头Adobe宣布，将在知名视频编辑软件Premiere Pro的新版本添加多款文生视频AI工具。近两个月震动全球科技圈的Sora，以及两款同类产品Gen-2和Pika，都会在不久的将来加入“Adobe全家桶”。

有了Sora等第三方AI工具的助力，新版Premiere Pro除了编辑处理预先拍摄的常规视频外，还能根据用户输入的文本即时生成AI视频，并将两者融为一体。

Adobe放出了一条官方演示视频：一个男人走向窗前，观看整个城市的夜景。用户无需拍摄实景，只需输入一段文字，即可利用Sora生成一段城市雨夜的视频，并与前面的视频无缝衔接，效果几可乱真。

Sora今年2月初次亮相，OpenAI放出几段演示视频，但并未公布产品进展和上线时间。如今，Sora被Adobe接纳，表明过去几个月又有不小进步，距离开放使用更近。

另一边，背靠抖音的视频剪辑软件剪映，同样朝着AIGC（人工智能生成内容）方向前行。但截至目前，剪映尚未拿出令人惊艳的成绩。

目前，剪映的AI玩法不少，包括一键成片、剪同款、AI克隆音色、数字人口播等，但并不具备根据文本直接生成视频的能力。它的海外版本Capcut在2月底推出文生视频功能，但效果距离Sora相去甚远。

如今，Sora除了在技术和产品上领先剪映，还得到了Adobe的青睐。Adobe在全球拥有超3300万付费用户；Sora接入“Adobe全家桶”，有望获取订阅收入分成，从而初步构建商业模式。

这也意味着，仍在打磨AIGC能力的剪映，追赶Sora的难度将越来越大。

做交互设计10年，我为何转岗到产品经理？

真正转岗之后，我发现很多工作还是超出了自己的想象。产品经理的工作确实比较杂。理论上，产品经理的工作包括了产品的方方面面，从市场研究、用户调研、数据分析...

查看详情 >

剪映被外界视为抖音搭上AIGC时代快车的关键筹码。得益于抖音的加持，剪映已成为用户量最大的手机视频剪辑软件之一，每天产出大量短视频；在此基础上更进一步，从UGC（用户生产内容）迈向AIGC，似乎水到渠成。

今年2月7日，抖音功勋老将张楠辞去集团CEO职务，亲自带队剪映。她在官宣职务变动的内部信中称，AI图像生成对她产生很大的触动，潜力巨大，并决定“放下一切”，义无反顾地出发。

剪映的发展根基不可谓不好，抖音的重视程度也很高。但两个多月过去，剪映依然没有太大动静，Sora反而继续高歌猛进。

在AIGC时代，抖音及其背后的字节，似乎总是棋慢一着。

字节八年前布局AI，成立了专门的实验室，招募大批业内精英，但成果却停留在内容审核、自动翻译、搜索服务等。2022年下半年，ChatGPT引爆行业；字节随即增加大模型投入，一口气推出AI聊天机器人等十几款应用，却迟迟没有拿出震动行业的技术和产品。

今年1月底的年度全员会上，字节CEO梁汝波感叹，字节“该有的大公司病全有了”。他特别点名AI业务，称：“公司层面的半年度技术回顾，直到2023年才开始讨论GPT，而业内做得比较好的大模型创业公司，都是在2018年至2021年创立的。”

如今，在立身之本的视频内容领域，OpenAI着着抢先，剪映乃至整个字节又双叒慢了。

01

背靠抖音这颗大树，剪映原本有很大机会在AI文生视频领域的占得先机。

剪映2019年上线，2021年推出PC端的专业版。它显著拉低了UGC视频的创作门槛，普通人只需准备好素材、点击几下，就能生成一段质量尚可的视频，并一键发布至抖音。

此外，作为一款工具软件，剪映还带有一定的社区属性。除了官方提供的视频创作课堂外，用户还可以参考达人制作的视频模板进行创作，也就是“剪同款”。这有助于提高用户留存率和活跃度。

依靠丰富的功能，以及与抖音深度绑定，剪映仅用了三年就获得超1亿月活跃用户，跃升至同类软件的第一名。

剪映海外版CapCut的增长同样惊人。

Capcut上线于2020年，与TikTok关联密切。移动应用分析平台点点数据显示，目前Capcut的月活跃用户超2亿。另据市场调研机构data.ai的数据，截至2023年8月，Capcut在iPhone和安卓端的用户超4.9亿，相当于TikTok全球用户量的1/4。

除了用户规模庞大，剪映在AI领域的另一个优势是“近水楼台先得月”，坐拥大量可用于AI大模型训练的视频数据。

数据、算法和算力是AI大模型的三大基础要素，其中数据是根基。要想提升大模型的综合能力，就需要不断“投喂”多模态数据，主要是互联网文本、图像和视频。

OpenAI、Adobe等公司都不直接掌握数据，必须从第三方付费获取，成本不菲。据媒体报道，OpenAI每年仅采购版权新闻文章授权的费用，就高达100万~500万美元；Adobe为了训练自家的文生视频大模型，以每分钟3美元的价格，向另一个大模型服务Midjourney购买视频片段。

相比之下，剪映在国内绑定抖音、在海外背靠TikTok，每天有大量用户使用它剪辑和上传视频。这让剪映能够以相对低廉的成本，接触到大量视频内容，为训练大模型、探索AI文生视频功能打下基础。

但从2019年至今，剪映并未研发出强大的AIGC能力，反而早早启动了商业化。

剪映很早就上线了VIP会员，用户每月支付二三十块钱，可以使用专属素材、精选模板等，AI玩法也被列为会员权益之一。根据官方介绍，AI玩法主要包括“无限创作”“无限运镜”和“瞬息宇宙”，基于现有视频素材进行AI美化编辑。

不难看出，剪映的AI玩法距离Sora这样的文生视频相去甚远。此外，它的收费也并不低廉：会员每月可获得1200积分，而使用一次“无限运镜”就需要480积分。积分耗尽后，用户可以选择继续充值，兑换比例为1元：100积分。

剪映诞生之初的定位是，尽可能拉低短视频创作门槛，促进抖音UGC生态的繁荣。过去五年，它的确完成了这一任务，大量抖音爆款视频背后，都有剪映提供技术和模板。特别是那些趣味特效视频、踩点视频，抖音红人发布之后，大量用户迅速群起效仿、共同推高热度；没有剪映的帮忙，恐怕不易做到这一点。

但如今看来，抖音还是把剪映的天花板设定地太低了。它原本有机会成为Sora这样的划时代产品，却始终停留在视频剪辑软件的范畴。

今年以来，剪映开始奋起直追，但先机已失、对手已至，追赶难度也迅速增大。剪映被Sora甩开，字节在AI领域仍然没能跳出“起大早、赶晚集”的怪圈。这也从侧面印证了梁汝波此前批评的“平庸的重力”。

02

过度重视对业务的帮助，过早追求商业化，是剪映乃至整个字节AI板块的难题。

剪映除了开设付费会员、将AI玩法纳入权益包，还在APP内增加了不少广告位。例如，用户点击“剪同款”，映入眼帘的除了手机照片和视频素材，还有悬浮在素材上方、几乎毫无关联的横幅广告。

作为一款用户破亿的工具软件，剪映的常规投入和资金压力并不会特别大。它之所以早早发展会员和广告，或许与字节的做事风格和评判标准有关。

众所周知，字节是一家超快节奏、高度内卷的互联网巨头。如果个人、团队或业务无法迅速带来肉眼可见的产出，就有可能被调整，甚至彻底出局。即使是需要长期投入的AI，也无法跳出这一隐形的评判标杆。

早在2016年，字节就设立了AI Lab人工智能实验室，并引入多位学界和行业精英。彼时，OpenAI同样初出茅庐，正朝着通用人工智能的愿景前行，将自己视为非营利组织。

相比之下，AI Lab虽然名为“实验室”，实际上仍然需要密切配合和服务业务。它的官网宣称，其研究重点是开发为字节内容平台服务的新技术；具体领域则包括自然语言处理、数据挖掘、计算机视觉、机器学习等，且和抖音各业务板块贴合紧密。

随后几年，字节拿出了一系列AI工具，比如提供翻译服务的Byte Translator，AI写稿机器人Xiaomingbot，以及头条和抖音的搜索服务等。它们固然颇有价值，但算不上突破AI领域认知边界、定义AI发展范式的创新产品。

直到2022年下半年，OpenAI多年磨一剑，ChatGPT席卷全球，生成式AI成为全球科技公司竞逐焦点，字节才把更多精力投入到这一新浪潮中。

已经淡出字节一线管理的张一鸣，对AI产生强烈兴趣，并鼓励团队大举投入。字节跳动无法错过AGI（通用人工智能），它是抖音和TikTok在全球发现新的增长机遇不可或缺的伙伴。

创始人发话后，字节各业务部门闻风而动。从那时起，字节陆续上线十多款AI产品，如豆包、话炉、扣子、Gauth等，剪映、飞书等也增添了AI功能。

但在这一轮大举投入中，字节以业务需求为原点、对标竞争对手的做事逻辑被延续下来。豆包等产品均为现有AI技术的场景化应用，而非对于AGI的原生探索。

例如，去年3月，微软上线整合了GPT功能的365 Copilot，全球办公软件市场为之震动。一个月后，飞书宣布即将上线AI助手“My AI”，以对话形式提供多种功能，包括优化和续写文字内容、创建日程、自动汇总会议纪要、搜索公司内部知识库等。

一年后，GPT已经在微软“全家桶”遍地开花，带动后者股价从250美元一路上涨至400美元以上。飞书却没能依靠My AI完成逆袭，反而在3月底宣布裁员。

又比如，Sora今年2月16日发布演示视频，CapCut一周后就宣布推出这一功能，每人每天可免费生成5段视频。CapCut的技术力显然无法与OpenAI比肩，其文生视频功能也较为简陋；仓促上线、对标Sora，不免带有蹭热点、强行完成KPI的味道。

从张一鸣点名AGI至今，字节AI又走过了一年，并未显著拉近与OpenAI的差距，甚至有扩大的趋势。字节AI向实用主义的过度倾斜，不仅让它错失了以往的机遇，也有可能拖累追赶的步伐。

03

对于AI板块各自为战、围绕业务打转的局面及其危害，字节已经有所察觉。

2023年11月，字节抽调多个部门的精兵强将，组建AI部门Flow。TikTok技术负责人朱文佳、字节产品与战略副总裁朱骏、字节技术副总裁洪定坤、飞书产品副总裁齐俊元均加入其中。人员的“高配”，显露了字节通过Flow统筹AI发展、消除重复建设的意图。

目前，字节曝光度最高的AI产品——豆包、扣子、话炉等，均由Flow部门负责。剪映、飞书、大力教育等虽然也有AI业务，但声量已经逐渐被Flow盖过。

另一方面，字节云雀大模型已经打磨近一年时间，多模态大模型BuboGPT也取得进展，为Flow批量产出AI应用奠定根基。假以时日，Flow有望扭转字节AI总是慢人一步的局面。

不过，字节毕竟不是OpenAI，没有微软这样的超级金主。AGI固然重要，但对于现有业务的拉动尚不明显，需要更长时间的沉淀和打磨，才能释放商业价值；字节对AGI的投入不可能无止境，必须考虑中短期的投入产出比。

短期来看，字节需要用钱、用人、用资源的地方很多，比如抖音货架电商、生活服务等，都需要大量真金白银。随之而来的现象是，尽管Flow挑起了大梁，但字节AI依然呈现兼顾业务需求的倾向。

据Tech星球近日报道，抖音生活服务刚刚成立一支AI团队，希望利用AI技术创造增量业务价值，且已开启对相关AI产品的研发，包括搭建生活服务相关的内容创作平台等。

抖音生活服务之所以拉起这样一支队伍，或许与竞争对手美团和饿了么的动作有关。

美团近日开始小规模测试AI助手服务“问小袋”，可为用户推荐符合其需求的外卖商品，以及用餐建议。饿了么则在4月初面向零售行业商家发布“AI经营助手”，可为商家智能生成各类经营关键报表和关键数据。

在此情况下，不让美团饿了么专美于前，或许是抖音生活服务入局AI的基本目标之一。

内外因素影响下，字节没有条件像OpenAI那样厚积薄发。既要追赶AGI潮流，又要具备快速落地、服务业务的能力，是字节AI的双重目标。

在张一鸣和梁汝波的督促下，外界无需质疑字节做AI的决心。但倘若再过去一年、两年，字节仍然跟不上OpenAI的节奏和水准，那么它或许应该考虑另一种选择：退回“卖水者”的角色，做AGI训练素材提供商。

如前所述，数据是大模型的三大基础要素之一，而字节麾下的今日头条、抖音、TikTok等，已经积累了数以亿计的文本、图片和视频。字节可以用这些数据训练自家大模型，也可以考虑更进一步，在妥善解决安全和隐私问题的前提下，将其出售给OpenAI等第三方公司。

充当AGI时代的“卖水者”，其实是字节擅长的流量生意的变体。字节如今的现金牛——广告和电商业务，都建立在流量变现的基础上；如果把AI公司转变为新客户，那么字节心心念念的第二增长曲线问题将迎刃而解。

另一方面，OpenAI既然能够与微软、Adobe合作，与字节合作也并非不可想象。毕竟，抖音和TikTok是Sora等AIGC服务最庞大的落地场景。倘若字节与OpenAI达成“竞合”关系，那么字节将跳出一步慢、步步慢的循环，在OpenAI的助力下搭上AI快车。

参考资料：

字母榜，《字节AI：失之PICO，收之飞书？》

字母榜，《Sora会“杀死”剪映吗？》

光锥智能，《不甘只做AI 应用工厂，字节跳动补课大模型》

Tech星球，《美团抖音下注新战场：内测外卖助手“问小袋”，成立生活服务AI》

新智元，《3300万剪辑师被革命， Sora、Pika、Gen-2将全面登陆Adobe》

撰文：彦飞，编辑：王靖

来源公众号：字母榜（ID：wujicaijing），让未来不止于大。

本文由人人都是产品经理合作媒体@字母榜 授权发布，未经许可，禁止转载。

题图来自Unsplash，基于CC0 协议

该文观点仅代表作者本人，人人都是产品经理平台仅提供信息存储空间服务。