Shulou(Shulou.com)12/24 Report--

Altman's play "between coming and going" in OpenAI ended with a return.

The process is very grasping, the onlookers are very excited, of course, the happiest is Microsoft.

Not only because no matter how this "farce" develops, Microsoft will win, but also because there is a bigger melon lurking behind it.

The biggest beneficiary of the big model is the unexpected veteran players like Microsoft.

For example, Stable Diffusion and Midjourney are scrambling to roll up the sky iteratively, but the one who makes a lot of money from AI mapping is actually Adobe.

Since the beginning of the year, the US stock market value of Adobe has risen by more than $1000, unwittingly creating an increase of 90.58 per cent, while Microsoft, OpenAI's largest shareholder, which launches a range of AI Copilot products, has also risen by a remarkable 55.93 per cent.

This kind of non-native AI veteran players, it seems that the main products are still very traditional, but AIGC products quietly occupy the mainstream-for example, Microsoft has just launched Windows AI Studio, where developers can access a variety of AI models and adapt to their own needs.

As soon as the news came out, I heard the wail of many startups on the track.

Why is the traditional company the first to harvest in the big model era?

To understand this, you must first understand that the industry has changed since the initial outbreak of AIGC technology.

The development of AIGC is roughly divided into two stages from the technology blowout to the market landing of the traditional company after AIGC.

In the first stage, AIGC technology blowout, startups and new products continue to emerge.

Both Adobe and Microsoft seem to play the role of background board in the first phase of the technology explosion. Its traditional products, such as PhotoShop and Office family buckets, may even become the first "old applications" to be revolutionized by AIGC technology.

But now that the first round of investment and financing is over, AIGC has entered the second phase of its search for landing, and instead of dying out, these traditional companies have become more active:

Not long ago, Adobe launched Firefly Image 2, which has a resolution up to 2k and has been fully incorporated into Photoshop, Illustrator, Premiere Pro and other suites. Microsoft released Copilot Studio based on the established office software Office and browser Bing.

△ Firefly Image 2 updated effect, image source Adobe

Both Adobe and Microsoft have gained a lot of attention relying on the new features of AIGC.

Startups, by contrast, are beginning to face landing challenges:

OpenAI is not yet profitable; Stability AI, the star startup that made Stable Diffusion, has been exposed to revenue difficulties, and some executives have recently left because of copyright issues.

Why do seemingly traditional companies like Adobe and Microsoft complete a round of "counter-killing" in the era of AIGC 2.0 and take the lead in profiting from new technologies?

On the face of it, these companies have just updated their products with the help of AIGC technology tuyere.

But in fact, they hold the most critical part of the product from technological landing to market matching--

Scene.

Take Adobe as an example, its paid product PS accounts for nearly 70 per cent of the design market, bringing extremely stable cash flow; coupled with the fact that many core users are commercial designers, buying Adobe members is the most convenient way to introduce AIGC to reduce costs and increase efficiency without changing the workflow.

This is because, from the point of view of users, such as designers, including program evaluation, delivery and modification of finished products, can not be separated from the communication and cooperation with other positions, this process involves a set of business flow.

Although other AI graphic tools such as DALL ·E 3 can generate a large number of pictures at a low cost, these images are neither freely editable vector graphics, nor do they necessarily meet the needs of users. In a word, they are not in line with the business flow and will bring additional workload.

However, as a product carrying business flow, Adobe has a better understanding of the pain points where designers really need to reduce their workload. After the introduction of tools such as generating vector graph (Firefly vector model) and AI map modification (Generative Fill), it is equivalent to directly using AIGC to simplify business flow and save working time.

On the other hand, Adobe and Microsoft, which have done a lot of work, only represent the way traditional foreign companies played in the AI 2.0 era.

By contrast, how are the traditional domestic companies doing?

In the past two days, Whale, a domestic digital service provider for marketing and sales, held an autumn conference with the theme of "AGI for Growth unleashes the power of AGI growth" related to technologies such as AI and large models.

As one of the representative companies in the field of domestic marketing, what kind of new products have been launched for AIGC? For marketing, and even for the whole AIGC industry, what experiences can be summed up from the cases of strategic planning?

Let's take it apart one by one and look at it in detail.

Head scene players, so using AIWhale strategy, domestic AI marketing and sales scene of the head players, 4 years of business (2022) to join the ranks of unicorns.

Then came the birth of ChatGPT, which is a new beginning for the marketing field and the whole industry. As a result, the business of this digital marketing operation platform, which focuses on AI technology, has become even more popular.

After the large model has been transferred from pure technical dividend to "win the world", enterprises with landing scenarios can combine AI and large models with business to form new engines and new business closed loops.

The advantage of the strategy is the marketing scene that it has precipitated for many years, in which there is not only a lot of business experience, vertical data, but also an understanding of the needs of the industry.

For the marketing industry, the big model brings the most useful capabilities, mainly two points, one is data summary, the other is content generation.

Among them, because the large model can help marketing and convey the breadth and depth of the content, content generation is more important.

Going back to the strategy we used to lift the chestnut 🌰 this time, how does it make the breadth and depth of the content generated by the AI general model show in the professional field of marketing?

First of all, a brief introduction to this conference, the strategy presented a whole set of products, including iterative and new release models, including--

"AGI Cloud Alivia", "Whale SpaceSight", "content Marketing Center Whale Harbor", "Whale Echo", "Whale Cast".

The lowest-level capabilities of these products are provided by Alivia, an enterprise-class AGI toolkit designed for marketing and sales.

It is called a toolkit because Alivia not only covers the marketing track, but also has everything from AI voice, AI posters, AI video editing to AI digital live streaming, and can also provide enterprise-level model training and management.

With the large model ability of Alivia, you can save time and effort to reach more users.

Take e-commerce now inseparable from the use of AIGC capabilities to produce promotional materials, for example, life diagrams, pictures and video editing are already necessary functions, if not satisfied, you can even train your own enterprise model.

The ability to generate text diagrams, the content can be accurate to letters, and the subjects described in black and white will not be confused at will:

If other scenes such as "back" are emphasized in the prompt, you can also accurately change the posture of the model:

As for the ability to create pictures, we need to go a step further, whether it's changing the background of the product in the poster:

Or replace a poster model:

For Alivia, it is a small case that is competent in seconds.

And advertisers can define the style and style of the desired poster according to the tone of the brand. On this basis, it is easy for AIGC to generate the required poster:

If you encounter a marketing scene where the Shanghai News cannot hold, you can also use the ability of a video editor to make a flash video and push it in front of you every minute, and it can be the kind that contains a digital human image:

Based on the large model capability foundation laid by Alivia, the strategy has launched a series of AI products, the core of which are as follows:

SpaceSight / math space, applied to digital offline store operation

Harbor / content Marketing Center, applied to digital content marketing

Echo / Picture, applied to digital voice service

Cast / broadcast, applied to the operation of digital live studio.

These four are the further manifestation of the ability of the big model of Alivia, so that the marketing content can be deeply transmitted.

The so-called depth, in fact, can be understood as a comprehensive and in-depth understanding of every industry and every user, so that when the generated content appears in front of the user, step on the customer journey transformation point as accurately as possible.

To put it more bluntly, everyone is exposed to content that they can understand and are interested in, so as to improve the marketing conversion rate.

To achieve this idea, in addition to the accumulation of user preference data from online (studio) and offline (stores), it is also inseparable from Harbor in content marketing.

Harbor collects and collates all kinds of data, including the speech skills collected during the live broadcast, the quality of the words, or the passenger flow and retention rates in different areas of the store.

What happens when the data summary ability of the large model is used, combined with the content generation ability to sort out and present the relationship between the data and the data, and then fed to the large model?

The fact is, with every operation of or sales, the work can be doge, and the boss can compare the commodity data in minutes.

A platform that makes good use of the ability to summarize large model data and generate content, and provides it to every customer who needs it, the impact is now directly accessible and perceived.

It's not just "working". Further, AIGC has formed an "amplifier effect":

Marketing in the usual sense is about using technology to help customers increase traffic and conversion rates, but now with the help of AIGC, it will only become faster, cheaper and more efficient.

But now that AI and big models are available to everyone, old companies and new startups seem to be on the same race, driven by technology. Why is it the veteran contestants such as Adobe and choreographer who make more profits? Are there any experiences to learn from?

Out of curiosity, we contacted the founder and CEO of the strategy, Ye Shengyi, and talked to him about the positioning of the strategy in the AIGC scene, what we think of the AIGC industry, and how AIGC will affect players in various industries.

Who is harvesting the AIGC technology dividend? Mr. Ye graduated from California Institute of Technology in Computing and Neurology, worked in data science at Facebook, and then returned to China to enter the digital marketing industry and founded Whale.

When it comes to AIGC, Ye Shengxing believes that the current players can be divided into four layers.

At the bottom are startups like OpenAI and Stability AI, or companies that own the technology AGI itself. This type of players have general large model technology, and will also provide the lowest level of technical support for higher-level players.

The third layer is the players who are more inclined to the infrastructure layer, such as LangChain and AgentChain, who will make some technical bases that support general large models, and many of them are open source tools to further support the development of AIGC and even large models.

The second layer is the further birth of AI tools products based on the first two layers of architecture, such as companies developed by technologies such as LoRA, which focus on providing the tools themselves and opening up small-scale innovative ideas.

At the top is the application company, including foreign companies such as Adobe, Microsoft, and domestic companies. This kind of companies have scenarios, and rely on the scenarios to create a lot of applications with stable traffic, but also expand the business to the first few scenarios based on demand.

Which of these four categories of players can harvest the final technical dividend?

Ye Shengfeng's view is:

Startups cannot reap the final technology dividend.

Take OpenAI as an example, although this kind of companies can rely on technological innovation to achieve their own value, quickly break the circle, and even make their own efforts to promote the development and growth of the whole AIGC technology ecosystem.

But in the end, the underlying technology such as the big model still needs to be supported by computing power, data and scenarios, so it will eventually be harvested by cloud vendors like Microsoft.

But further, scene players who can harvest technology dividends do not rely solely on the scene itself, that is, products that have been tested in the market.

Ye Shengfeng believes that there are mainly two aspects of standards.

On the one hand, these companies should have the ability to mine the value of AIGC technology and understand the different needs of different users for algorithms from the scene.

Take image generation as an example, although the type of this classical algorithm is limited, the algorithm details of the requirements are actually very different for customers in different scenarios.

For example, for the household industry, whether the design details can be constructed perfectly and reasonably when generating furniture, the algorithm must be constantly polished in the vertical field in order to make it an "expert designer":

Then take the scene understanding as an example. For brands in different industries, there are also differences in the "rules" of operating offline stores, such as food safety standards in food stores and commodity arrangement requirements in clothing stores.

How to create a scene understanding algorithm that conforms to the enterprise based on these differences in details also requires access:

Therefore, no matter which field it is, the practical application of AIGC technology will be different. How to use the appropriate technology to bring the greatest technical blessing to their own scene is a problem that scene players need to think about.

On the other hand, the company itself must have technical strength in order to quickly follow up with the arrival of a new wave of AIGC and maximize the value brought by technology to the scene.

Ye Shengyi said that as a technology company, Chuanzuo will also inherit the open source culture of Facebook, actively embrace open source, contribute to open source, and invest in AIGC technology research.

For example, the strategy has also contributed to the open source community of AI, such as Stable Diffusion, and participated in the follow-up of various projects, which is not contradictory to the development of the product, but complement each other.

Of course, no matter what kind of players, in the end, they have to judge the wind direction of AIGC from the present and the future, and choose the appropriate landing ideas.

After all, for AIGC technology itself, some people still hold a cautious attitude, thinking that it is suspected of being over-touted, that the existing AI generation ability is not perfect, and there is still a certain distance from being directly available.

Ye Shengfu gave a summary of AIGC's current technological progress and whether it is worth investing in the future.

The development and progress of technology itself will not be blocked, just as AGI should not be a goal or technical criterion, but a path to meet the needs of end customers.

As for judging whether a technology is worth investing, it is ultimately necessary to return to three indicators, that is, whether the technology landing is really useful, whether the market is large enough (for example, counterpart translation is not a big market), and whether customers are willing to pay the bill.

Therefore, instead of waiting for the arrival of a certain node, we should find our own scene and grasp the technological path in order to harvest the final technological dividend in the competitive battle.

This article comes from the official account of Wechat: quantum bit (ID:QbitAI), author: Hengyu Xiao Xiao

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.