As one of the large model unicorns, the "hot" of the dark side of the month continues. While there are rumors of investment by Tencent, recently, Ali also revealed that it has invested about 800 million US dollars in the dark side of the month. The dark side of the moon, which is crazily sucking money in the primary market, is also exploring the commercial closed loop in the field of to C by way of reward. Combined with the current hot price war, the commercial anxiety of the big model is self-evident.

Suck gold

Ali is using practical actions to prove the core position of the "user first, AI-driven" strategy, and the dark side of the month, as the darling of capital, is bound to be within Ali's consideration. Recently, Ali financial report revealed that the company has invested a total of about $800 million in the dark side of the month in fiscal year 2024, buying about 36% of the shares, which is the preferred stock investment of the dark side of the month.

Cai Chongxin, chairman of Alibaba Group, and Wu Yongming, CEO, wrote in a letter to shareholders that "in the next decade, no industry will be immune from the subversion brought about by AI."AI as a driving force for groundbreaking user experience and business model will bring great opportunities."JackpotWe are excited about the unlimited possibilities of AI driving the 'user first' strategy.

According to the collation of public information, since the second half of last year, Ali has accelerated the speed of investment in the field of AI. Before the dark side of the month, it has successively invested in four large domestic model enterprises, namely Zhisu AI, Baichuan Intelligence, 010 million things and MiniMax, almost all of them are rising rapidly and becoming rookies in the unicorn industry.

When Ali publicly invested in the dark side of the month, the market also heard news that the dark side of the month completed a new round of financing, including Tencent, Gaorong Venture Capital and other institutions. After a new round of financing, the dark side of the month's valuation has reached $3 billion.

It is worth mentioning that Tencent's investment objectives and Ali have a high degree of overlap, in addition to MiniMax, wisdom spectrum AI, Baichuan intelligence, but also includes deep talk technology and unasked core dome and so on.

According to SkyEye check information, three months ago, the dark side of the month just completed its last round of financing of more than US $1 billion, with a corresponding valuation of US $2.5 billion, making it one of the leading companies in the field of large models in China.

At that time, it was reported that this is also the largest round of financing for domestic large model companies, including Sequoia Capital China, ZhenFund and so on. Up to now, the dark side of the month has made four rounds of investment, including Meituan, Lanchi Venture Capital and other well-known enterprises.

As soon as the news of Tencent's investment came out, some analysts believe that Tencent mainly hopes to establish a deeper cooperation between Wechat and the intelligent assistant of Moon's dark side product Kimi. Beijing Business Daily contacted Tencent and the dark side of the month for Tencent's investment and the possibility of Wechat's cooperation with Kimi. As of the month of publication, Tencent did not respond, while Tencent said it would "not comment".

Cash out

The Dark side of Moon was founded in 2023, and the core members of the founding team participated in the research and development of many large models, such as Google Gemini, Google Bard, Pangu NLP, Wudao and so on. In March this year, the long text model kimi launched by the dark side of Moon exploded "out of the circle", which also detonated the long text track of the big model.

"the dark side of the moon does have some unique things on the large model, such as its own unique direction in model training, rather than the casing commonly used in other large models, and the latter's training accuracy is usually not satisfactory." in an interview with a reporter from Beijing Business Daily, Shen Meng, director of Xiangsong Capital, said that from the founder's technical, engineering background and other aspects. Although the dark side of the moon has not yet appeared the core technological innovation, it is indeed one of the few enterprises in China to build a big model.

According to public information, Yue Zi Dark founder Yang Zhilin, with a bachelor's degree in computer science from Tsinghua University and a doctorate in computer science from Carnegie Mellon University, studied under Apple's former head of AI, once worked at Facebook AI Research and Google Brain, and is the most highly cited researcher in natural language processing under the age of 35 in China.

Yang Zhilin is known as the representative of the technological belief school in the domestic scientific and technological circles, but this does not mean that Yang Zhilin is going to "generate electricity for love" all the time, just like the dark side of the moon when there is no shortage of money, nor can he never consider commercial realization.

A few days ago, Kimi tested a reward function in a small range of grayscale, with an amount ranging from 5.Jackpot.2Mel 399 yuan, which can guarantee the priority of paying users when there is a lack of computing power during peak hours.

At a time when Kimi has not yet become a real popular style application, the reward model of the dark side of Moon has also triggered a discussion about whether it is premature to explore the commercialization of large models. Shen Meng's analysis said that the big model itself is very expensive, and if the enterprise can not find a direction in technology for a long time, the future development space will be greatly limited, which will affect the patience of investors. this forces enterprises to get commercial benefits as soon as possible.

"large model enterprises have generally been able to do it in the same way as heap parameters, and the commercialization we are seeing now is partly because we have entered a stage of bubble and homogenization competition, and those who commercialize earlier will be able to recover some of their costs earlier," says Shen Meng.

Low price

The dark side of the month's fiery financing and careful exploration of commercialization have a bigger background. For some time, byte, Baidu, Ali, Tencent and other big companies have taken the lead in price reduction, and the big model has made a "sharp turn". The application of popular style has not yet appeared, and the price war has come first.

Ma Qianli, co-founder of Unbounded AI, told Beijing Business Daily that the price reduction of the big model is mainly focused on the reasoning side, mainly for three reasons: lower computing costs due to a more adequate supply of computing power, a market trend derived from fierce competition, and the opportunity to build their own ecology in order to obtain more users and more data to optimize and iterate their own products.

With regard to the price war of the big model, the market is also divided into two emotions: the "optimists" believe that the decline in prices will promote the penetration and popularization of artificial intelligence technology in various industries and promote the real prosperity of the AI ecology, while the "pessimists" believe that the root of the price war is homogenized competition, and the outcome of crazy price reduction will be a lose-lose situation.

Ma Qianli is a member of the optimists. He believes that the price war reflects that the whole industry still has very high expectations for the development of large models, and it will also lead the outside world to pay more attention to AI, which is equivalent to more definite expectations. Ma Qianli added that the current price war is more about the competition of general models, while the vertical domain model is relatively better.

However, Ma Qianli also mentioned that the staff of large factories he had contacted generally showed an anxious mentality."This anxiety does come from homogeneous competition. Although there are gaps between models, many times it does not show a crushing advantage. Enterprises all have their own KPIs, and if they invest so much, it will inevitably lead to 'rolling the market', otherwise they will face a situation where the KPIs cannot be completed."

Big manufacturers such as Tencent and Alibaba, while having their own big models, are also actively investing in other big model companies, which may also confirm this anxiety that spreads between industries. Shen Meng said that the current big factories do not have many core advantages in terms of large models. In this case, they rely on their financial strength to run around as much as possible and participate in investment in large models with characteristics or well-known, to a certain extent. It is also sharing risks.

Beijing Business Daily reporter Yang Yuehan