Source: Titanium Media

Author | Lin Zhijia

Image Source: Generated by Wujie AI

"We are against 'internal competition' and 'open source'. Blindly open sourcing will make the AI field very chaotic, and it is very easy to let the industry be dominated by inferior products, which is not a healthy industry landscape. I have always believed that in the first quarter of 2024, the industry pattern of (large models) will basically settle, and more than 100 large models will become undifferentiated," said Li Di, CEO of Xiaobing, to Titanium Media App recently.

In 2023, AI large model technology represented by ChatGPT has become one of the most discussed topics in the tech industry, with hundreds of millions of people directly interacting with generative AI tools.

However, contrary to the majority's enthusiasm for "large models leading a new industrial revolution" and "AI reshaping the world," Li Di is a true innovator of AI Beings who opposes the consensus, opposes the "competition" of large models and computing power, and proposes alternative views.

"I am indeed not a believer in 'large models'," Li Di told Titanium Media App.

Li Di has stated that AI technology innovation is currently in a period of turbulence, with some large model products in an unverifiable state, suitable business models for the industry have not been established, commercialization faces challenges such as low-price vicious competition and the inability to generate high-profit value despite cost reduction, and has fallen into a "vicious circle." The Xiaobing team is actively seeking new business models with more value.

On January 4th of this year, Xiaobing announced that it has obtained domestic filing for "Xiaobing Large Model" and officially launched multiple technical products such as Rinna large model, Xiaobing clone, Xiaobing large model digital employees, and has realized the landing and commercial development of self-developed large model products with different parameter scales and uses. At the same time, some top internet celebrities use Xiaobing clones to provide digital content, with personal monthly income exceeding 100,000 yuan and annual income reaching 1 million yuan, achieving preliminary commercial validation.

Recently, Titanium Media App conducted an exclusive interview with Li Di for nearly 90 minutes, discussing the development changes in China's AI large model industry over the past year, Xiaobing's development model overseas, future trends in AI technology development, and the relationship between AI and humans.

Li Di stated that Xiaobing has always hoped to create AI partners that can establish long-term emotional bonds with people, and continues to explore unique paths of innovation. This year, the company will increase technological iteration and scale commercial validation, and bring more original AI Native products to the industry. He emphasized that AI applications like Xiaobing, which provide emotional companionship and are interesting, have longer-term commercial value than useful AI products.

Helping internet celebrities earn millions, Xiaobing's team opposes the "internal competition" of large models

Xiaobing's predecessor was the Xiaobing team of Microsoft (Asia) Internet Engineering Institute, which was the largest independent AI product development team of Microsoft globally, established in China in December 2013. In 2020, Microsoft announced the spin-off of Xiaobing as an independent entity and continued to maintain investment rights.

In November 2020, Xiaobing completed Pre-A round of financing. On November 24th, Microsoft China and Xiaobing announced a strategic partnership in Beijing, joining forces to launch a series of AI + cloud computing commercial solutions for key industry customers.

On May 16, 2023, Xiaobing announced the launch of the "GPT clone program," recruiting the first batch of 300 people. Today, Xiaobing has developed multiple technical products such as "Xiaobing clone" and "Xiaobing digital employees."

Xiaobing's image photo (Source: provided by the interviewee)

Looking back on more than three years since the spin-off, Li Di has deep feelings. He told Titanium Media App that entrepreneurship in the field of AI technology is not easy, with too many temptations and too little action. The AI field is by no means a "winner takes all" game, but requires a long-term, continuous cultivation and business exploration process.

"The influence of the exam-oriented education in China is too great. Only two types of people can give true advice: one is a clear 'successful person,' and Xiaobing is not one; the other is a clear 'failure,' who can share their lessons with everyone before dying, and Xiaobing is not that kind of person," Li Di said. Many people are not suitable for starting an AI company, and one reason is that entrepreneurs influenced by external investors will make the company overly "capitalized."

Therefore, unlike the aggressive financing strategies adopted by other companies, Xiaobing's last public financing occurred in 2022, with a scale of 1 billion yuan. Important investors include Netease, Jiyuan Capital, and other institutions.

"Objectively speaking, we are not 'arrogant,' but we do not want to 'chase' in the AI bubble, and do not want to spend more time on this," Li Di told Titanium Media App. "Investors are not your lovers or your partners. The investment behavior of VCs is their evaluation of input and output, and they have independent value judgments."

Li Di and his Xiaobing team present a unique attitude in the current environment: not catering, not participating in "internal competition," not competing in computing power, opposing the "open source" of large models, and also opposing the "accumulation" of high computing costs.

Of course, this does not mean that Xiaobing does not develop "large models." Li Di stated that as early as 2021, Xiaobing had developed multiple large model technologies, and even earlier, Xiaobing had created chat products like "Xiaobing," successfully enabling thousands of users to communicate. Xiaobing has always been deeply cultivating "AI Beings" - AI-driven emotional interactive digital humans, and has developed products such as Xiaobing clones.

Li Di admitted that chatbot products are no longer the most interesting to users. "Unfortunately, it's already in the past. For any chat, or rather, conversational experience, most people don't have such a strong feeling anymore. Why is ChatGPT popular? Because it is useful," Li Di said.

"Since 2018, Xiaobing's entire framework has been emphasizing her comprehensive ability. The problem is not how she talks, but whether she has a 'circle of friends,' can communicate and provide feedback, which is the decisive factor for the future development of AI Beings," Li Di told Titanium Media App. AI needs to evolve into emotional expression, making it easier to establish a user "connection." AI clones not only need to be used, but also need to generate a sense of "possession" in people's minds.

Currently, the Xiaobing team has developed "Xiaobing clones," where creators can clone themselves through Xiaobing's framework technology after identity authentication and release them. Therefore, the clones have the creator's personality, memories, knowledge, voice, and appearance, and can freely converse, generate photos and videos, and form group lives. After half a year of testing, Xiaobing clones have attracted 800,000 creators, and top creators are expected to exceed a million in income, and users can interact directly with clones on Taobao.

In addition to the clone program, in terms of AI singers, Xiaobing has cooperated with Netease Cloud to launch AI music creation software, and Luo Tianyi and others will join; in terms of digital employees, new products such as digital interactive business cards have been launched, and live streaming solutions have further expanded to support minority languages and assist cross-border e-commerce; Xiaobing in third-party platforms such as Xiaoai and OPPO is gradually transitioning to large models; in addition, Xiaobing's Rinna large model has achieved top rankings in multiple professional lists in Japan.

Li Di admitted that whether it is Xiaobing clones or Xiaobing large models, they will not be "open source."

"Open sourcing will make many things in the AI field very chaotic, and it is very easy for 'bad money to drive out good money.' Without the authorization of the individual, there is a high risk of 'issues.' For example, AI Sun Yanzi, if the individual does not authorize, the legitimate rights and interests of many people will be harmed. Although the AI open source agreement is relatively 'vague,' we are not inclined to continue in a 'vague' manner," Li Di pointed out. "Once you enter the 'deep water area,' you will find that the open sourcing of models does not have a significant impact, but at the same time, the continuous high cost of model training computing power makes it difficult to 'earn profits.'"

Li Di emphasized that the AI that Xiaobing's team wants to create should not only be useful but also have "warmth," allowing emotional AI digital humans to change humanity. He hopes that in the next step, humans and digital humans can come together to form a new social network. "I believe that this way, she can also have a more sustainable development."

"In fact, in the past period, some have exaggerated or overstated (AI large model technology) facts," Li Di believes that AI large models are not a race, nor are they an industry, and everyone has some misconceptions.

In his view, AI large models are just a "basic capability," relatively expensive, but not so scarce, and do not have a high technical threshold.

Many people have imagined that AI could disrupt all industries, but in the end, it did not become a reality. For example, Meiya Camera did not disrupt the "hippocampus," and AI large models did not disrupt traditional industrial systems.

"First, at the beginning of 2023, everyone thought that the larger the model parameter scale, the better, but that's not the case; second, many people thought that through models with huge parameters, capabilities would continue to 'emerge,' and AGI (Artificial General Intelligence) would arrive soon, but this does not conform to scientific laws, so we chose to do 'hybrid models'; third, many people thought that computing power would rapidly decrease, and business models would be established as computing power decreases, but we judge that computing power costs will not decrease, and computing power costs will remain high for a long time," Li Di said. These are all gaps between fantasy and reality.

Therefore, Li Di believes that AI large models cannot disrupt all traditional industries; they are just an auxiliary tool.

He pointed out that the current AI large model track is too "internally competitive," which can easily cause confusion, and the industry may experience the phenomenon of "bad money driving out good money." Li Di predicts that after the "big waves wash the sand" process, the market pattern of the AI large model industry may settle by the first quarter of 2024.

Regarding the industry trends in 2024, Li Di summarized three points: first, computing power costs will not decrease, and data training will continue; second, more large model companies will exit the market, and it will also be difficult for large model companies to merge, "merging two large models is meaningless"; third, AI Beings, this type of digital human model, will also see commercialization landing, for example, on January 9th, Xiaobing and DingTalk announced a comprehensive cooperation in the fields of short videos, live streaming, video conferencing, and AI assistants.

Li Di admitted that Xiaobing will not directly sell large model APIs by "compressing costs and reducing prices," but will pursue "a packaged product form." At the same time, Xiaobing aims to provide digital human products with high added value to make customers feel the true "value." "I believe that everyone should do something meaningful, rather than 'rolling' models."

Li Di once cited that currently, in more than 60% of the top ten automobile brands in China, the conversational systems in intelligent car cabins are provided by Xiaobing. Additionally, 90% of financial text generation in China is also developed by Xiaobing. He believes that Xiaobing will continue to be a leading player in domestic AI competitions.

Li Di emphasized that large models are only a part of the company's many technological iterations, and Xiaobing will not become China's ChatGPT. In addition, Xiaobing has a complete AI technology framework and insists on technical originality and independence in key technologies.

Regarding OpenAI's recent technical research, Li Di stated, "Frankly, I believe that OpenAI's Q* research exists, and many people are also researching in this direction, but has he produced any research? Currently, it is 'non-existent.'"

Li Di emphasized that we all hope to live in a magnificent and prosperous era, but reality is cruel, and there is still a long way to go for the commercialization of AI technology, and it will take longer to verify the widespread application of generative AI technology.

"I don't know why we always have a 'time waits for no one' mentality. When large models appeared, it seemed that everyone thought this would propel the entire industry forward; experts even warned that if you don't learn AI now, you may not find a job in the future; companies also fear falling behind their competitors, afraid that if they don't chase, they will 'fall behind' and lose at the starting line. However, the reality is not like that. In my opinion, the current situation is like playing the Minesweeper game, where everyone is trying and judging in their own area, but cannot accurately determine whether the next tile is a 'mine' or a 'safe square.' When a new area is uncovered, everyone will suddenly realize and have an understanding. This is the essence of the world - the unknown we face always exceeds the known future," Li Di said.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。