As 2023 drew to a close, a seismic shift took place in the realm of artificial intelligence, propelled notably by large language models and generative AI technologiesThis surge catalyzed the emergence of the "battle of the hundred models" in the Chinese market, where nearly 200 generative AI models had completed regulatory approval and gone live by July of this yearAdhering to the principle that "great efforts yield extraordinary results," mainstream AI models began a competitive race towards an expansive increase in parameter size, aiming at the ambitious goal of achieving Artificial General Intelligence (AGI) in the near futureHowever, entering 2024, numerous obstacles - be it computational power limitations, funding issues, or otherwise - rendered AGI still a distant realityConsequently, the industry initiated a "demystification" process regarding large models, prompting a quest for genuine transformation of these models into a new form of productive capacity that can effectively unlock commercial potential.

The second half of this "battle of the hundred models" is expected to be particularly captivating

What developmental characteristics will Chinese large models exhibit throughout 2024? How will the landscape of domestically produced models evolve? What progress is being made in terms of industrial implementation and commercialization? A recent series of interviews and investigations have aimed to unravel these pressing questions.

A pronounced "Matthew Effect" is becoming increasingly evident in the industryLarge models, particularly foundational general models, operate under a high investment model that necessitates substantial computational power, vast datasets, and robust algorithmsThe major players in the landscape primarily consist of three categoriesThe first category includes tech behemoths like Baidu, Alibaba, Tencent, ByteDance, and Huawei, who have significantly increased their investment in large models as of 2024. For instance, Tencent's Mix Yuan model launched a video generation feature this month, boasting 13 billion parameters and full open-sourcing

• U.S. Treasuries Face Pressure Amid Rate Cut Debate
• Google's Chip Launch Drives Market Cap Surge
• Top Five Hottest Stocks of the Year in U.S. Market
• Fed Rate Cuts Spark Market Collapse
• Price War Engulfs the Bottled Water Sector

Baidu’s Wenxin model has experienced a dramatic rise in daily average token usage, exceeding 150 million by November 2024, a staggering 30-fold increase from the 50 million tokens reported a year prior, while its user count reached an impressive 70 millionByteDance has also revelled in success, reporting over 160 million cumulative users for its Doubao model and maintaining a stable daily increase of 800,000 new users.

The second category features emerging star AI startups such as Zhipu AI, Dark Side of the Moon, MiniMax, Leap Star, and Wall of MindThese companies are often helmed by founders with distinguished backgrounds and have caught the attention of investors, successfully closing multiple funding rounds with market valuations reaching the 20 billion yuan markNotably, Zhipu AI clinched 3 billion yuan in new financing on December 17. Official disclosures reveal that the company’s commercialization revenue has surpassed 100%, with its C-end product "Qingyan" app serving over 25 million users and annual revenues over 10 million yuan.

The third category comprises research institutes and universities, such as the Beijing Academy of Artificial Intelligence, Shanghai Artificial Intelligence Institute, Tsinghua University, and the Chinese Academy of Sciences

The Matthew Effect experienced by domestically produced large models is further illustrated by recent evaluation results released by the Intelligent Origin Research InstituteOn December 19, the institute published and interpreted comprehensive and specialized evaluation results from over 100 open-source and commercial closed-source models across various domains including language processing, audiovisual capabilities, and moreAmong the top five models featured in these assessments, the likes of ByteDance’s Doubao model, Alibaba’s Qianwen model, and Zhipu AI were prominently recognized alongside well-known foreign entities like OpenAI and Anthropic.

Interestingly, a price war has emerged in the domestic model market, starting from the first half of 2024 and sustaining momentum into the later monthsByteDance ignited the latest phase of this price competition, announcing on December 18 that the input price for its Doubao visual understanding model is set at merely 0.003 yuan per 1,000 tokens

This essentially allows users to process up to 284 images of 720 pixels per inch for just one yuan, offering an 85% discount compared to standard industry pricingEarlier in May and June, Chinese firms led a notable price-cutting spree, with ByteDance, Alibaba, Baidu, Tencent, Zhipu AI, iFLYTEK, and others inflating the competitive atmosphereZhipu AI's CEO, Zhang Peng, responded to inquiries by asserting that their strategic approach to commercialization transcends a mere "price war," emphasizing a commitment to iterative innovations in core model technologies and enhancements in efficiency to achieve sustained cost reductions for applications and continuous value upgrades for clients.

A noteworthy observation by academic experts indicates that the foundational large AI models are progressing from text-based unidimensionality towards multimodal capabilities encompassing text interaction, image creation, and video generation

This evolution hints at a deeper integration of AI across various industries, propelling a rapid transformation of sectors towards smarter, more value-driven paradigms.

As the competition heats up, applying these models in various industries has become a central focusThe initial hype surrounding general large models led many to believe that they would revolutionize customer engagement and the software sector - a perspective rooted strictly in technological advancementsHowever, the practical application of AI large models should center on real-world industry challengesRecently, Kong Miao, the Vice President of Ronglian Cloud and founder of Zhuge Intelligence, pointed out to reporters that there has been a noticeable shift in the wind for large models throughout 2023. In the beginning, enterprises were preoccupied with purchasing model capabilities, yet by the second quarter, demand for real-world applications began to swell.

In fintech, large models have penetrated over 50% of the market share, leading the way among various sectors

Yet, such penetration doesn't translate to improved productivity levelsLast year's trend saw financial institutions focusing primarily on training specialized models built on foundational large modelsToday, however, the spotlight has shifted entirely to application-focused solutions encompassing diverse financial services such as smart customer service, agent assistance, securities quality control, and digital marketing, all requiring highly tailored solutions to enhance operational efficiency.

Regardless of the industry, the shifting demand reflects an overarching trend—from a generalized model approach towards more defined applicationsThis metamorphosis speaks to enterprises' genuine need for effective product enhancements and higher returns on investment (ROI).

He Jun, the Vice President of TCL Industries and CEO of Gechuang Dongzhi, emphasized that the integration of AI with industrial software is a pivotal technological turning point towards the smart factory of the future

This fusion presents vast imaginations, aiming for a seamless connection between internal data and external ecosystems to satisfy individual user needs.

According to Wang Aihua, the Deputy Chief Engineer of the China Academy of Information and Communications Technology, China stands firmly among the global AI development eliteCurrent statistics illustrate that the country houses over 4,500 core AI enterprises, fostering a well-rounded industrial base and comprehensive application service capabilityPresently, China is ardently promoting an "AI + Action" strategy, encouraging the strengthening of AI capabilities in fostering industrial modernizationIn the industrial sector, applications of smaller-scale AI models have become increasingly commonplace, with insights from 507 cases indicating that quality management, equipment maintenance, and operational management scenarios dominate industrial applications, accounting for nearly half of the assessed instances