Picture source: Generated by Unbounded AI

"I haven't been idle this Spring Festival." More than one executive of an AI company sighed like this after the start of construction in 2025.

China's AI industry is accelerating at an unprecedented speed.

DeepSeek not only bombarded the US technology circle during the Spring Festival, but also forced the entire Chinese AI industry to catch up with progress around the Spring Festival - to issue models, accept products, and open source.

"New products are released in the middle of the night" has become the new normal for Chinese AI companies: whether it is the recently popular Manus or Tongyi's latest open source reasoning model QWQ-32B, "when you wake up, you are bombarded by dark horse companies and new technologies."

The industry that changes three times a day often bombards the nerves of AI practitioners.

Although these "dark horse" companies are famous for their "Hangzhou Six Little Dragons", the area with the most AI companies is still within 5 kilometers of Wudaokou, the "Center of the Universe" - this is the sphere of influence of the Tsinghua Department and the Institute of Automation of the Chinese Academy of Sciences, and the two universities that cultivate the most top AI talents in China.

In order to recruit talents from Tsinghua University, DeepSeek founder Liang Wenfeng placed his Beijing office in Rongke Building about 2 kilometers south of Tsinghua University; the company behind the intelligent Manus is hidden in the office park area of ​​Huayuan Road Community, Haidian District, Beijing, adjacent to ByteDance.

The former "cosmic center" has restored its glory with the blessing of the big model.

Around a crossroad at Wudaokou, there are many AI star startups such as Zhipu, Baichuan Intelligent, Shengma Technology, Wall-facing Intelligent, Wuwenxinqiong, and Trekjing Technology.

Iron office buildings, flowing startups. When the big model develops ten times faster than the Internet, every company is running wildly, otherwise it may not be able to survive the "Zaocao King" of Wudaokou.

Every cafe here, you may bump into an AI practitioner; after a party in the evening, someone will turn back to the company and continue working overtime until the early morning.

The excitement of opportunity and the anxiety of inclination are the smell of air here.

The first quarter of 2025 is about to pass, and the popularity and reshuffle of AI companies are happening, and the ranking of products in various test sets is constantly changing.

2025 is the year when the industry expects Agent (agent) explosion, and it may also be the year when the battle for basic big model ushers in the end.

In the short term, technological advantages will be the best weapon to win the time window. During this period, a tug-of-war against scenarios, traffic, and talents is unfolding. Challengers with more resources and defenders with the first-mover advantage are both running desperately.

Expected adjustment: The battle of the fundamental model ends, vertical competition intensifies

If you use one word to connect 2024 and 2025, the "expected adjustment" is appropriate.

OneYears have passed, and the king is no longer there. OpenAI, which was once in its peak, is gradually surpassed by its competitor Anthropic. The latest released big model GPT-4.5 no longer brings sensation to the industry, and it is more disappointing. New technologies and products are still emerging at high frequency, which arouses the emotions of practitioners.

Even Zhu Xiaohu, who did not believe in AGI and refused to see any AGI company last year, said, "I will definitely vote" after seeing DeepSeek.

Repeated subversion of cognition frequently occurs to the heads of AI companies.

When mentioning the keywords of AI development in the past year, Kong Miao, Vice President of Ronglian Cloud and founder of Zhuge Intelligent, told Guangma Intelligent that the adjustment of expectations for the market value of large models in 2024 will be from fully buying in model capabilities to starting to pay attention to applications, from technology-driven to business participation.

"Whether it is adjusting technology or planning for commercialization, everyone's expectations are being adjusted," said Kong Miao.

Scientist Ilya, who left OpenAI, announced that "pre-training has reached the upper limit", the public data used for training has long been exhausted, and the rumored GPT-5 did not arrive as expected, and the Zero-10, one of the six little tigers in China announced that it would give up super-large model pre-training, and then DeepSeek emerged and broke the shackles of computing power.

In the first two months of the start of 2025, changes in the AI ​​circle have caused an earthquake at home and abroad.

For DeepSeek, which was popular during the Spring Festival, why can its emergence inspire practitioners? What significance will DeepSeek bring to the industry?

Pixel Bloom PixelBloom (AiPPT.cn) Founder and CEO Zhao Chong summarized it into three points: the technical dividends brought by open source, the cost threshold for C-end AI applications has been reduced, and China's AI "scenario definition technology" has entered a new stage.

Zhao Chong said that DeepSeek's open strategy forces the entire industry to rethink the boundaries of technology monopoly. In addition, the cost revolution it triggers will determine the end of business.

"When the inference cost of the 100 billion parameter model dropped from 'driven supercar' to 'riding shared bicycles', it means that AI applications can finally serve ordinary users on a large scale," said Zhao Chong.

On this basis, a number of products will also usher in new possibilities for "scenario definition technology" - whoever can solve more specific problems with lower costs will be able to re-format the rules of the game. This is also the battlefield that Chinese entrepreneurs are best at.

In addition, DeepSeek also accelerated the final war for the fundamental model. Several people told Guangma Intelligent that the battle of basic model is expected to come to an end in 2025 and the dust is settled.

Since the beginning of the year, Zero-1000 has announced the withdrawal of super-large model pre-training, frankly announcing the company's transformation from pursuing AGI to focusing on commercialization. Changes in several other companies are also continuing: MiniMax is released for the first timeThe open source model has been used to replace the traditional transformer architecture with the "linear attention" mechanism; Baichuan Intelligent released its first full-scene inference model Baichuan-M1-preview after the New Year, and Wang Xiaochuan is still focusing on the application in the medical field.

Kaifu Lee once mentioned that "super-large models are games that only large companies can play", and for the other five companies, the huge war of burning money and saving talents is also difficult to sustain.

But after DeepSeek became popular, the story of big models relying on computing power seemed to be rewritten again, and the resources of big companies seemed to have lost to the ultimate technological innovation.

Tang Jiayu, co-founder and president of Shengshu Technology, saw that a group of companies with technological advantages had the opportunity to overtake. He told Guangma Intelligent that from DeepSeek, it can be clearly seen that technology has changed from computing power dependence to algorithm dependence, which means that the computing power resources owned by large manufacturers in the past are no longer obvious advantages. A group of large-scale model startups with core technology thresholds will use their own algorithms and other technical advantages to quickly occupy the market.

In contrast, most people are more optimistic about companies that focus on B-end business for the next echelon changes in Liu Xiaohu.

Many industry insiders told Guangma Intelligent that in this war of big model startups, to C's company may end the war earlier, while to B's company survived longer.

"Relative to the C-end, the B-end actually has a defensive depth, and it is a place where large companies cannot make it purely by traffic." Zhao Chong told Guangma Intelligent that it is difficult for to C companies to resist the competitive pressure of Byte "Doubao" and Tencent "Yuanbao", but for companies such as Zhipu that have a group of B-end customers, these are reliable, and the B-end is also more suitable for early commercialization.

In addition to the competition of general big models, competition in some vertical industries that are becoming more mature is also intensifying.

Taking AI videos as an example, in this track where players such as Sora, Keling, Shengnu, Conch, etc. have evolved from the original PPT form to a more realistic quality.

As for the increasingly "volume" AI video generation track development, Tang Jiayu's judgment is that this war will come to an end this year.

"This year, there may only be 3 top companies in the domestic AI video generation field, and 5 companies in the world."

In Tang Jiayu's eyes, in 2025, AI video generation has reached a stage of "everyone can use". For novices who have no experience in making videos, they can also easily get started.

Behind the "everyone can use it", is that the "impossible triangle" of AI video is being gradually broken, that is, speed, cost and quality can be achieved at the same time.

In Zhao Chong's view, the domestic war between AiPPT.cn was completed as early as 2024, "just consolidate its advantages." Next, their focus will be on overseas markets.

If generative AI is likeA sandwich cookie, in addition to the basic model and application layer, AI Infra in the middle has always been considered a business with strong certainty but not so sexy.

Lin Song, an investor who focuses on the AI ​​hardware field, told Guangma Intelligence that in 2025, the AI ​​Infra field will be more "volume".

"These companies can at least survive, and they can also have retail and profits, but if the industry cannot find some particularly large growth space, I think the growth rate may decrease rapidly, and it will be more uncomfortable once it is rolled up at this time." Lin Song said.

Lin Song said that after experiencing the exponential growth stage in 2023, the slowdown in investment in 2024 is a normal phenomenon, both because of the stable competitive situation and a process of de-bubble. "In 2023, most of the companies invested from 0 to 1, so they must be investing heavily. The later they go, the funds (proportion) the company needs will also decrease."

Spend money and complete teams, and the Chinese giants will indeed dance. In this gradually tightening war, the big factory team is investing more manpower and financial resources. All in AI.

Free, open source, and linkage DeepSeek are all appetizers. Whenever opportunities in new sub-sectors appear, major manufacturers are extremely determined to get on the bus.

After the emergence of new phenomenal products, there must be large manufacturers among the anxious group of people.

Xiang Guangzhi Intelligent, the head of AI products of a large factory, said that on the afternoon of the day when Manus detonated the AI ​​circle, his group specially held a meeting to urgently discuss: "How did Manus achieve it?" and "How long will it take for us to reappear as soon as possible."

In addition to chasing technology, major manufacturers are making advance arrangements with scenarios and ecology, and adjustments are happening at any time, and the changes in rankings may only be overnight.

This adjustment is first reflected in the repeatedly changing organizational structure of major companies. From Byte to Alibaba and then to Tencent, the big model R&D team and to C product team were split into one step of the game.

Ali’s AI to C strategy is becoming clearer day by day. From the end of 2024, the Tongyi App team belonging to Alibaba Cloud was merged into Alibaba Intelligent Information Division, and then this year, Quark and Tongyi Qianwen completed the merger.

A person close to Alibaba commented on Guangma Intelligently that the Quark team played fiercely. By giving the Tongyi Qianwen team, which is weak in commercialization and stronger engineers, and relying on Quark's 40 million DAU, the traffic can be directly imported to the latter.

This year, Tencent has also accelerated the pace of splitting the team. In the past two months, Tencent has adjusted Tencent Yuanbao from the TEG business group to CSIG (cloud and smart industry business group); transferred product teams such as QQ browser, Sogou input method, and ima from PCG (platform and content business group) to CSIG (cloud and smart industry business group), and put together products upgraded through AI.

One of which, Tencent Yuanbao was also handed over to Tencent AssociationThe person in charge of Wu Zurong is responsible. Tencent may hope to rely on the person in charge with rich experience in C-end products to enhance the influence of Tencent Yuanbao.

If there are still major manufacturers in the waiting and testing stage in 2024, by 2025, all major manufacturers are betting on the possibility of the birth of super AI applications and calling for talents and money at all costs.

"In the first year of starting, it is normal that the team has not been adjusted well," an industry insider told Guangma Intelligence. "From the first half of 2025, the bosses of each company will select the strongest players to lead the team and re-arrange the troops."

For example, Tencent, which has been slow in AI in the past two years, has former Tencent's technical middle-level analysis to Guangma Intelligence. Tencent has not applied it well to the development of AI in the accumulation of talents and experience after the arrival of the big model. For example, the Tencent Youtu team, which previously focused on machine learning, did not take over the training of AI video models, and it is difficult for existing personnel to quickly switch directions when they turn to the new visual model architecture direction.

In addition to adjusting the team, spending a lot of money is also the norm for big companies. Compared with startups, large companies with sufficient capital pools have more investments in the hardware side.

After the year, the future plans announced by several major factories have revealed the same ambitions. On January 23, Reuters revealed that ByteDance will allocate more than $20 billion this year for AI chips, data centers and other hardware. On February 24, Alibaba announced that it will invest more than 380 billion yuan in the next three years to build cloud and AI hardware facilities, with the total amount exceeding the total in the past decade; Baidu announced the next day that it will invest 11 billion yuan in AI infrastructure.

But in the AI ​​2.0 era where uncertainty is stronger, real money may not necessarily bring miracles, and a reshuffle will happen at any time.

In 2025, known as the "year of AI Agent", ecology is not only a hope for large factories to overtake, but also a barrier that small factories can't reach.

Some people use DeepSeek to promote their own products, while others turn the bow in time under the inspiration of fresh blood and move towards the road to open source.

In the past two years, Robin Li, who firmly believed that "remaining the source can lead to better commercialization", began to actively embrace open source. Not only will its flagship model Wen Xin Yiyan 4.5 be announced to be open source on June 30, but it has also decided to supply all models under Wen Xin Yiyan for free.

"What I have learned in the past few months is that open source can help you get more attention. We are in the early stages of AI and generative AI innovation. Faster dissemination will help improve adoption, but it will also help more people try this technology, thereby enabling innovation at the application layer." At the World Governments Summit 2025 summit, Robin Li talked about open source like this.

From the open source of text models, this "open source wind" has also continued in the multimodal field. On February 25, Alibaba, the "open source major" once again opened the Wanxiang 2.1 video productionModel; on March 6, Tencent once again opened the map video model based on the previous open source Wensheng video model.

At present, among the four highly anticipated major manufacturers, Alibaba, Tencent and Baidu have all firmly taken the path of open source. Compared with the first three companies, ByteDance, which is committed to building an "AI application factory", is still closed.

In making basic models, Bytes is still recruiting soldiers and may have higher expectations for internal self-developed models.

On February 17, Dr. Wu Yonghui, who works at Google Gemini, joined ByteDance. It is reported that he will serve as the head of basic research of Seed, a big model team, focusing on basic research of big model.

From Zhang Yiming staying up late to read papers, pulling authors to chat, to adjusting the department early, dispatching troops, and setting up Flow, a department responsible for AI applications and Seed, who is in charge of big model technology, it can be seen that Byte's determination to bet on AI is getting stronger and stronger day by day.

A person close to bytes told Light Cone Intelligence that bytes has developed faster in the Q4 stage than in the previous few months. This is because ByteDance's overall company's strategic height has been further improved, and the various resources it invests continue to "increase".

Following the "violence of miracles" style, Byte lifted the bean bun to the top 1 among similar products. However, this position is not a worry-free place.

"From the perspective of Doubao, its first competitor is Tencent, followed by Baidu and Quark," an interviewee told Guangma Intelligent, "However, the potential biggest threat is Huawei. With the advantages of smart terminals, Huawei will become the most powerful competitor."

Confrontation or symbiosis, how to choose a startup company?

Startups are not inferior to giants in the "volume".

As time goes by, the pressure on startups is increasing. Faced with large companies with traffic and scene barriers, should startups confront or symbiosis?

In the second year leading to AGI, many startups have fallen out of financing and not enough money. The best outcome among them may be to be acquired by a major manufacturer, such as Character.AI, which was incorporated by Google. What's worse is that the company declared bankruptcy and employees are still in arbitration, trying to get back the owed wages.

But in this round of development of the AI ​​industry, the acquisition methods for startups have changed. The acquirer no longer generously accepts all the employees and property of a startup company, but only takes away core talents and computing power equipment. When the core of a company is hollowed out, all that remains is the shell.

In the increasingly difficult entrepreneurial environment, "not touching the business of large companies" has become a tacit principle for entrepreneurs.

From OpenAI to DeepSeek, technological innovation is still the most effective way to turn the tables of major manufacturers.

Review the differences between the AI ​​2.0 era and the AI ​​1.0 era,As a witness to the two eras, Tang Jiayu could clearly feel the similarities and differences between the two.

"The two eras rely on accumulation, but the AI ​​1.0 era relies more on the accumulation or acquisition of mature experience, such as clearly labeling data in the solution and mature talents with experience in model parameter adjustment. As long as the resources are in place, things can basically be done." Tang Jiayu said, "But the AI ​​2.0 era has more uncertainty, and is in the development period of technology, and the path to solving technical problems is not certain. Startups also need to proceed step by step based on their own cognition."

But because of this, startups have the opportunity to overtake.

AI companies in parks near Wudaokou

However, this generation of entrepreneurs is different from 2014. At the beginning of their business, few companies could not consider the possible wars with large factories. For startups, how to survive is also a matter that must be considered at the beginning.

For AiPPT.cn, which is more inclined to use applications, they established their own plan to coexist with big manufacturers on the first day: through cooperation, they seize as many traffic portals as possible with big manufacturers, and become the number one player in the vertical track.

"Our positioning is not a general agent, but a vertical agent, so we basically have a cooperative relationship with all major manufacturers." Zhao Chong said, "At present, about 30 core major manufacturers in China, more than half of which choose to cooperate exclusively with AiPPT.cn. In this area, we have almost no competition in China and are almost the first in the gap."

Through linkage with major manufacturers platforms, smart hardware and other channels, Zhao Chong said that he can always get free traffic. "For example, Lenovo's AIPC shipments increased by 30% to Q2 last year, so I just follow it." The advantage of this is that it does not cost too much money to invest in investment, but also reaches as many users as possible.

Zhao Chong shared that before the end of 2024, AiPPT.cn has successfully exceeded the 10 million users mark.

At the user growth level, AiPPT.cn's idea also highlights the idea of ​​avoiding competition with major manufacturers.

"We have to make misalignment with big manufacturers. Before we make products (products), we will communicate with big manufacturers first to avoid those products with high frequency and urgent needs." Zhao Chong said, "PPT is actually a low frequency and urgent need. In this way, we can choose to complement large manufacturers as much as possible."

In determining the selection of the track, some startups are also following the idea of ​​avoiding them. Among them, some companies choose small-scale markets, which are not within the range of large manufacturers; others choose to chew on those hard bones that are more laborious for large manufacturers.

"The market size is too large, and basically all the major manufacturers are within the range of the range. The strategic departments of major manufacturers are actually very smart, so don't consider those who can make a lot of money." Zhao Chong said, "When the industry exchanges, friends share their experiences, and they won't do anything with more than 1 billion markets."

Focus on finance and customer serviceRonglianyun of the domain is doing the work of "chewing hard bones". In Kong Miao's eyes, the new orders brought by the development of the big model still need to be handed over to industry service manufacturers to complete.

"If a big model manufacturer bids, he will subcontract the work related to the industry application scenarios when he gets it." Kong Miao explained that because corporate service requires the accumulation of industry experience, large manufacturers generally choose to hand it over to service providers in vertical fields.

"The big model capability only solves the AI ​​generalization capability of many scenarios, but it only accounts for 20% of all demands, and the remaining 80% of the dirty and tiring work is good, but from the perspective of an industry application manufacturer, it may be a product standard." Kong Miao said, "The major manufacturers cannot do these engineering projects and technical needs of customers in the industry at a glance. In fact, it is not that they do not have the technical ability to do it, but that they do not have knowledge (industry professional knowledge), so they are not able to do such a detailed market."

The business lines of large factories are partially overlapping with the business lines of large factories. The practice of Shengma Technology is to open the time window through technical advantages, and then occupy as much of the user's mind as possible from the subdivided fields.

"Coexistence is a very likely thing to happen." Tang Jiayu said. For Shengma Technology, while constantly exploring its technical capabilities, it is also making efforts to implement the application of B-end and C-ends.

"From the strategic perspective, we will first gain a foothold in some sub-application industries for AI video generation, such as pan-entertainment, animation, etc., to provide B-end services to some platforms for C-end consumers." Tang Jiayu said. "For example, when users mention AI video generation animation, the first choice for users in China and abroad, including Japan is Shengma Technology Vidu. We hope to occupy more industry tracks like this and become the 'first choice' for users."

Take animation as an example, only a creative team needs to draw a few keyframes, and some of the middle clips can be handed over to AI to generate, which will greatly reduce the production cost and time. For example, in the production process of animated series, the general cost is about 100,000 yuan per minute, but Vidu can reduce the cost to less than one-tenth of the original while achieving the same video effect.

"The official promotional film of the movie "Venom" we undertaken before is an ink-wash version of the animation, which helped the production team reduce the post-production time by about 90%," said Tang Jiayu.

At the C-end level, Tang Jiayu shared that they are currently paying attention to the new content consumption model spawned by new technologies. With the advent of the AI ​​era, these new content may not be the most suitable for growing on Douyin and Kuaishou. In the evolution of content form, new opportunities are being born.

Only by finding your own positioning and planning, start-ups can run further and longer in 2025.

2025, the "year of explosion" of the agent

Users' perception of AI is becoming more and more obvious.

When family members around you can open bean bags to ask questions anytime and anywhere; when more jobs of writing reports, papers, and doing PPT are habitually thrown away by peopleGiven AI, no one will question whether AI is a bubble, but is constantly trying and exploring in the overwhelming new products.

Open the recent App Store app rankings, four of the top seven applications in the free list are AI applications. DeepSeek ranked first, followed by ByteDance, Tencent Yuanbao and Ali Quark.

Since 2024, a group of C-end application players of "100 billion clubs" have appeared. According to the AI ​​product list statistics, as of February, the 11 products led by DeepSeek have exceeded 10 million visits.

Many industry insiders told Guangma Intelligent that 2025 will be the "year of explosion" for intelligent bodies.

Just in early March, the emergence and popularity of Manus were a powerful symbol of the "Year of Intelligents" in 2025.

"Less structure, more intelligence" Manus, who shouted the slogan, proves that the overflow of large-scale model capabilities has reached a critical point. With sufficient support for the underlying architecture capabilities, how to connect capabilities together so that the agent can make as few mistakes as possible in the planning and execution process, this tests the product strength of AI companies.

Different from the past, other catching up companies need time to reproduce or even surpass the same technology, which often takes longer, but this time, the reproduction is completed in less than one day.

Whether it is the MetaGPT that opens "Open Manus" in 3 hours, or the OWL that has the best reproduction performance on the market. Their existence proves that before the vent came, someone had already done the same thing in the same direction.

"Agents that complete a single task will definitely be realized this year." Zhu Zheqing, founder of Pokke AI, said in a sharing session of Jinqiu Fund.

A group of apps that have upgraded or newly created with the help of AI capabilities will also usher in explosion this year.

From the perspective of necessary conditions - cost, thanks to the optimization of the cost of large-model bases, some AI application manufacturers have long achieved a balance of revenue and expenditure. Even if they do not make paid products, they can rely on the CPC (payment for browsing advertising) model to equalize costs and achieve profitability.

Wang Dengke, a developer who has made the "Cooling Simulator", recently shared that his team's AI companion application "Duji" has basically reached a balance of income and expenditure while free users are still available. "We don't burn money for big models and can support the team."

Some industries are being rewritten or even subverted by AI. In terms of the SaaS industry, Microsoft CEO Satya Nadella once made a prediction: AI Agent will fundamentally change the definition of SaaS, and it will even end some of the original service model.

"After the emergence of the big model, it has been applied in some industriesThe acceleration after the land was accelerated. We originally thought it was 5-10 years, but now it is 3-5 years. Many corporate service companies will definitely be killed. "Kong Miao said. In the past, big models played a role in assisting manpower, but now, agents can complete some multi-business process automation processes and then realize multi-agent collaboration, which is a big disruption.

In the process of combining AI and SaaS, at the beginning, AI will first increase competitiveness for enterprises through the ability of big models, thereby increasing customer unit price and increasing incremental volume. Taking customer service as an example, it was originally just a simple online robot, but now it can make some customer communication records and summary, and enterprises can charge separately for this service.

But in the medium term, this part of the growing market will shrink sooner or later. Kong Miao told Guangma Intelligence that by using tools to make most of the improvement, (Ronglian Cloud) will replace the original SaaS software process. "We no longer need trainers and business process configuration experts in the backend, but automatically configure the process through the big model, so the original software services will be replaced by end-to-end proxy software, eroding the original market. ”

In addition, replacing manpower by Agent can greatly improve the human-efficiency ratio, which is equivalent to turning the original software and services into real software services, allowing users to truly pay for the results.

For example, in the financial field, when securities companies need to respond to policies and include quality inspections within their business scope, the intervention of large models helps these companies save human resources.

However, due to industry knowledge barriers and the special requirements for privacy and accuracy in some industries, it still takes time to implement AI in thousands of industries.

"The integration of AI in the financial industry is inevitable, but the implementation of some needs needs to be solved, which is a matter of time. "Kong Miao said. "The current location of finance requires business processes and data tuning processes, and the implementation of software engineering RAG takes time. From the time it was impossible to scale at all, to now, it takes time to help the industry reshape business processes. ”

However, there are still some tracks that have not yet reached a clear turning point, and we need to wait for the technology to mature and the team to find differentiated advantages.

As for AI glasses, which are popular in the AI ​​hardware track, Lin Song believes that this track is still more advantageous.

"Earphones and glasses are closely related to mobile phones. If there is a mobile phone ecosystem, supply chain and industry users' cognition, user channels, and content connection, in comparison (mobile phone manufacturers) will definitely have more advantages. "Lin Song said, "As a new startup, you must have new creativity and a certain moat to break through the siege and avoid being quickly broken by large companies. ”

In the context of certainty and uncertainty, as Tang Jiayu said, in the era of AI 2.0, new possibilities are still emerging, and teams at the forefront cannot be just experience-dependent. AI companies need to accelerate their running by improving talent density and human-efficiency ratio.

Zhao Chong also shared his feelings with Guangma Intelligent:

"Looking back on these three years, there may be people who don't believe in AI in 2023. In 2024, basically everyone believes in AI and invests in it. In 2025, all apps will basically be transformed by AI, and all industries will be clear, and everyone is competing for execution." Zhao Chong said that in the past year, his team has doubled.

The group speed in the AI ​​era is just like another microcosm of the Internet era.