Undeniably, after more than 200 days of revelry, the big model circuit has come to the situation of "parameter in-roll, packaging concept".

The big manufacturers have used the big model as a business card to show the investment market that they are developing new business, and startup AI companies are using the open source big model as a benchmark to build big models in various branches of the track, trying to get a share of the pie. Of course, in this, there are also a lot of fish in troubled waters to rub the concept of players.

According to incomplete statistics, at present, the domestic large model of more than 1 billion parameter scale has been released 79, and less than this scale level is more. In a long period of time, the big model of the launch of the day as a unit in all parts of the country. The keynote speaker on the stage depicts a wonderful and promising story, leading the investors under the stage to fantasize about walking in the forefront of the times by means of big models.

The "Hundred Models War" is already on the verge.

In the face of this situation of almost obsession with big models, as one of the first domestic enterprises to realize the commercialization of AIGC, Xiaobing is particularly calm.

"In fact, now the industry should not be fanatical about what kind of application to make, but what kind of innovation and practical application the big model should have." Li Di, CEO of Xiaobing, said in a conversation with AI Blue Media Exchange.

In his view, if the big model can not let the AI company to get benign economic benefits from it, can only rely on free to get customers, broaden the market scale, is not to win the market with innovation, then such a business model will only be a flash in the pan.

"But unfortunately, the current domestic situation is so." He said.

Although not completely agree with this boom, as the CEO of an AI company, Li Di also still has his own thinking about the application and development of large models. At the same time, he also showed AI Blue Media Hub the innovation and layout of Xiaobing in recent years under the wind of artificial intelligence, including the birth of the clone AI Beings based on the concept of virtual idol, as well as with the social platform APP "small Iceland" and so on.

Perhaps, from his point of view, we can restore the real thoughts of the domestic veteran AI company about the big model and the AI wind mouth.

"What restricts the development of large models is the law of economics."

In February of this year, when many large domestic manufacturers still take a wait-and-see attitude towards large models, Xiaobing has tested its own large model product "Xiaobing chain".

However, in Li Di's words, the launch of the small ice chain is not so much a product release, but rather a strategic consideration of the company, but also in order to show the market the "decision-making" ability of the big model.

In fact, in the domestic large model of more than 200 days, "what can the large model" has become a common question in the minds of many people. Throughout the large model market, whether it is a large factory or start-up AI companies, the release of the large model of the use of more human-computer interaction and dialog and information retrieval, but it is clear that this is often hundreds of millions of dollars with the large model of the parameter scale and investment capital, as well as the market expectations do not match.

In this regard, Li Di believes that the role of the big model actually has three directions, respectively, decision-making, work and talk.

The latter two are well understood, the large model for the compression and extraction of large amounts of information can provide users with sufficient information, but also according to the reference material to summarize the generation of content, that is, the ability to produce content.

But in his opinion, the most important but least observable aspect of the big model is the ability to make decisions and logical judgments.

Li Di gave an example: on the "Tsinghua University and Peking University, which is better" this problem, in the past, we will be the original to find, and now ask the GPT it will be from the faculty, the number of national disciplines and even the number of men and women than the number of dimensions to be presented, which means that the big model can be used as a control center of the system.

"In the meantime, the big model will actually integrate its own emotional tendencies and logical analysis into each dimension, thus allowing the user to understand each other's intentions," he said, "Knowing that emotional intelligence itself is a rational ability, and people with high emotional intelligence are often imperceptible to others. "

However, Li Di has relatively clear thinking about the functionality of the big model, but for the question of the business model of the big model, he failed to give a relatively sustainable answer.

A fact is that now the application of large models of the ability to optimize and enhance human productivity, to provide creative inspiration for human beings, far from being able to achieve the level of substitution, and does not have the "uniqueness" of the enterprise is not willing to pay for the big model of an important factor.

He told a story to AI Blue Media Hub.

Some time ago, a media friend asked him how to produce a piece of media in line with the tone and quality of the article, he simply calculated an account, according to the general market price, every 1,000 words is worth 0.2 U.S. dollars, a regular 2,000-word article is worth 0.4 U.S. dollars, and it costs 40 U.S. dollars in total to produce 100 pieces of articles per month.

For a contributor, the fee would start at $1,500 for one manuscript, or $150,000 for 100 manuscripts.

He went on to explain to the other side: But I believe that the media will not pay 10% of the labor costs for AI-generated articles, and in the end, you will find that such a deal AI companies can earn 500,000 yuan is not bad.

Obviously, this figure is completely unable to match the economic benefits for the AI company, and whether it is by way of cost reduction and efficiency or free in-roll, it is difficult to change the underlying business logic.

"So, what restricts the development of big models and AI is actually the laws of economics, which is probably the destiny of AI companies."

Virtual idols + digital employees: two directions for AI development

Despite concerns, as an AI company in its time, Ice can't stop.

In Li Di's mind, there are only two directions of AI development in the future, one is "digital employee" or AI being, and the other is "digital clone".

The former has let most people in the market formed a consensus. Like two months ago, Zhou Hongyi said at the World Internet Conference, he believes that the popularity of the big model should be positioned as an assistant to enterprises and employees, through the "auxiliary mode" as a co-pilot role to provide assistance, so that the human will play a key role in the entire decision-making circuit.

This coincides with Li Di's view.

He has said that Xiaobing let various digital employees to work in various industries, similar to "labor dispatch", and finally get the salary, equivalent to the company's share of proceeds, and now the average annual salary of a digital employee can reach 300,000 yuan.

This means that the market has mostly recognized this business model.

As for the concept of digital clones, Li Di deliberately said that it is actually similar to the concept of virtual idol. This is also one of the main directions that Xiaobing is currently working on.

With Luo Tianyi such a virtual idol is different, Xiao Bing's digital clone is actually the reality of real people cloned into the AI to become a virtual star, such as the previous launch of the "Hanzou Forest" clone and so on.

"The influence of the Netflix community cannot be underestimated, and he himself carries a distinct labeling, like emotional bloggers and dressing bloggers. The advantage of digital clones is that they can communicate with users more often and are online 24/7, and their influence expands." Li Di said. "And, in order not to become like AI Sun Yanzhi and Sun Yanzhi's relationship, only the blogger's own initiative to clone will work."

At the same time, Li Di also mentioned that this system of digital cloning, the future may also be added to Ice's human and AI fusion of social products "small Iceland".

"Little Iceland to build a society in fact, the central thrust of the AI around the human, forming a virtual society similar to the animal friends." He said, "Letting AI serve people and recharging AI with emotional intelligence is actually a common starting point that AI companies should have when developing products."

Only, this product did not win too much attention from the market and C-end users. SevenMac data shows that it is currently ranked 153 on the social list, and the average number of downloads in the last three months is 752.

This may mean that the idea of a symbiosis between humans and AI is still a paradox for the time being, both on mobile apps and in daily life.

And when it comes to the relationship between AI and human beings, Li Di seems more relaxed.

"Just kidding, in fact, the only thing that really eats the dividends of AI now is phone fraud, and the technology it requires is far from the latest technology. So whether it's big models or AI, it's far from threatening human jobs at the moment."