Founder of Fengping Intelligence, Lin Hongxiang: AI digital player playback revenue exceeds 100 million GMV per day

Wen/Yan YanRecently, the entry for "Du Hua AI People Hua Hua Zi Live" hit the hot search. Du Hua, the digital avatar of Lehua Entertainment founder Du Hua in a light pink suit, "Hua Hua Zi", can chat with Du Hua himself in the live broadcast room, and in just over ten minutes, the audience exceeded 320000

Wen/Yan Yan

Recently, the entry for "Du Hua AI People Hua Hua Zi Live" hit the hot search. Du Hua, the digital avatar of Lehua Entertainment founder Du Hua in a light pink suit, "Hua Hua Zi", can chat with Du Hua himself in the live broadcast room, and in just over ten minutes, the audience exceeded 320000.

Huahuazi "is produced by Fengping Intelligent, an AIGC enterprise located in Beijing. Lin Hongxiang, the founder and CEO of Fengping Intelligent, told NetEase Technology that his team has achieved a business volume far exceeding the total of the past few years in more than six months, and the number of customers every month is almost doubling in scale. The company has now entered a profitable stage, and "a large number of VCs have come to us

From working in the artificial intelligence and big data team at IBM in the early stages to entering the underlying framework of designing big data algorithms at companies such as Baidu, Lin Hongxiang has gone through the entire process of global AI, from expert systems to machine learning to deep learning of big models. In 2014, he founded the internet industry insurance technology company "Da Te Bao" with a valuation of over 1 billion yuan and sold it to a Fortune Global 500 company.

Three years ago, sensing the opportunity for the AIGC industry to explode, Lin Hongxiang once again started his own business to establish Fengping Intelligence. His core ability is to provide customers from different industries with different needs with an AI model that is infinitely close to real people, allowing them to have a "digital person" that can replace real people.

At present, Fengping Intelligent has built an XGen digital human intelligence manufacturing system to produce intelligent digital humans with low cost and high quality, and has developed an original digital human AIGCSaaS tool to empower customers to achieve independent creation of digital human content. Compared with mainstream solutions in the market, it reduces costs by 90% and improves efficiency by 500%

We generally say that a 10 times increase in efficiency can transform an industry, but in certain specific fields, we have found that digital people can bring a 100 times increase in efficiency to the industry. The application has truly reached a critical point, and we can imagine the enormous productivity of digital people in the future, "said Lin Hongxiang.

Question: What impact has the sudden explosion of the AIGC concept brought to Fengping Intelligence?

Lin Hongxiang:Two years ago, when I said I wanted to become a digital person, many technical colleagues thought it was very unbelievable that we were doing something outrageous. Even in July and August last year, some investors told me that they wanted to give me a sum of money, but the premise was to work in another industry because they thought it was particularly unreliable.

But with the popularity of the AIGC concept, in the past six months, we have achieved business volume that has not been touched by our efforts in the past few years. The number of customers every month is almost doubling in size, and the company has now entered a profitable stage.

Question: Can you start from your frontline experience and share some scenarios where you can leverage digital people for rapid growth?

Lin Hongxiang:We can truly see the efficiency improvement brought by digital people to many industries. For example, we are serving a healthcare consulting client who has to pay over one billion related fees per platform every year. Now, we have made him a health consultant that is almost the same as a real doctor, and we have made him a special consulting brain, so that customized digital people can have the ability of in-depth consultation. It is estimated that their costs will be reduced by at least 80% this year, that is, 80 million yuan will be saved.

In addition, there is the education industry of renowned teachers, and the cost of good teachers is very high. Taking one-on-one courses with renowned teachers in Beijing may cost several thousand yuan per hour, and these famous teachers usually need to queue up for one or two months. We collaborated with a leading educational institution to clone the voice and image of famous teachers, and then trained their teaching and research experience into a model of famous teachers. This can meet the needs of "famous teachers" one-on-one answering different students, allowing good teachers who lack skills in the physical world to truly become skilled and meet the needs of more people.

Question: In your opinion, what are the new opportunities for commercial landing on the digital race track?

Lin Hongxiang:Fengping Intelligence focuses on the direction of AI digital people, which in simple terms allows digital people to have both a beautiful appearance and an interesting soul. The biggest scenario for the landing of AI digital human business is likely to be the digital employees of enterprises. Taking Fengping Intelligent as an example, the "No.1 AI Plan" has been launched, aiming to help each enterprise create the No.1 employee who understands the company the most besides the CEO, and help each expert create a clone who understands the expert's business the most besides themselves.

More specifically, I believe it is to help Chinese companies go overseas in this direction. When overseas companies are promoting overseas, the biggest problem they face is the extremely high cost of video advertising. We invite foreign models to shoot a video, which costs at least $500 to $1200 per episode. However, advertising is not included, and we spend $50000 on 100 episodes, which is too expensive.

Generally, regular advertising materials should account for about 8% of the cost, but cross-border e-commerce currently accounts for about 20%. If digital people are used, the cost can be reduced to one tenth of the previous level, and it can be produced for about $50. This is also an industry that we particularly hope to provide in-depth services to help Chinese enterprises better obtain more orders in the international market, cross cultural, time difference, language barriers, and other barriers, and achieve localized communication and communication through AI digital people driven by big models.

Question: How much real market demand is there for the current development of the digital human industry?

Lin Hongxiang:Taking us as an example, within one day of the beginning of the month, the AI digital people provided by Fengping Intelligence had a GMV of over 100 million in both public and private domains, which only included data from top partners willing to participate in the statistics. The digital human business has covered more than ten industry fields, including virtual customer service, live streaming sales, entertainment interaction, finance and insurance, culture and tourism, education, public services, etc. We have applied it in every dimension. With technological progress and scene expansion, it has entered a thousand industries, and there is still a lot of market demand.

Question: The complex collection of training sample data in the past has led to a long and costly customization process for digital humans, which to some extent limits the rapid application and implementation of digital humans. How does Fengping Intelligent achieve cost reduction and efficiency increase in training data collection, production costs, and other aspects?

Lin Hongxiang:We also made many mistakes throughout the entire process. Originally, our path was algorithm driven, followed by facial synthesis, voice synthesis, and brain synthesis. Each step of this process involved collecting a large number of samples. Deep modeling still requires each person to record for over forty hours, and then perform deep slicing. The overall customization process is very long, and the corresponding cost will remain high.

At the end of 2020, we continuously trained and iterated through deep neural network technology using the idea of large models and small samples. We validated over 140 technical paths and invested a small amount of 100 million yuan to develop our own digital human AIGC technology framework. Through two years of continuous iteration and repeated practice in the vertical industry, we formed an industry Know how solution, which achieved cost reduction and efficiency improvement, and crossed the critical point of commercialization. In the field of language big models, Fengping Intelligent is one of the earliest companies in China to release the FPLLM of the Fengping big model and apply it in the writing field. Since 2021, it has provided AI article automation creation services for domestic financial and insurance enterprises. At that time, fees were charged based on the quantity, quality, or diversion effect of articles, making it one of the earliest commercial big language models.

Question: What are the competitiveness and barriers of Fengping Intelligent reflected in?

Lin Hongxiang:Firstly, it is talent. We have an elite team from around the world in building talent for the big model. Another advantage is the first mover advantage, which accumulates large customer resources earlier than peers, and runs through many application scenarios in advance. In actual user needs, it accumulates accurate and high-quality data to meet more model iterations, gradually improving competitiveness, and forming its own growth flywheel.

But back to the technology itself, back to the algorithm itself, objectively speaking, there are no absolute barriers in this industry, and technology is likely to be surpassed. The most important thing for us at this stage is the gradual integration of different fragmented links, such as products, technologies, customers, and data. By building a growth flywheel for data, algorithms, and users through applications that satisfy customers, it is the current advantage of Fengping.

Question: After the rise of the AIGC concept, digital human enterprises have sprung up like mushrooms after rain. Which of these are "occupying the pit", which have real talent and learning, and what should be our judgment criteria?

Lin Hongxiang:I think there are two points, one is quality. Currently, there is a mix of good and bad in the market. On the supply side, many digital people provide low quality content, and a large amount of low-end, inefficient, low-quality, and repetitive content will lower user expectations. In the future, it is a challenge to produce digital humans with lower costs and higher quality, with an appearance that is no different from that of a real person, perfectly reproducing their movements and expressions.

Another thing is actually security. AIGC is a very big double-edged sword, which can generate huge value when used in good places, and can also become an accomplice to bad people when used in bad places. From our own perspective, we need to make AI safe and controllable. So we have built our own safety system, which can prompt for prohibited words during the production process. This management system can clearly see who produced what non compliant content at what time, and the content involved in violations will be manually reviewed and reviewed. After the content is released, we can even achieve tracking and traceability to ensure safety traceability. Only in this way can digital people truly become partners and service providers of real humanity, rather than competitors or harmers.