The large model market has thrived in 2024, marked by a shift in focus. On one hand, leading vendors have turned their attention to specific industry scenarios, accelerating the commercialization of tailored large models across sectors like finance, healthcare, and education. These models are progressively addressing intricate professional challenges. On the other hand, conversations around the future trajectory of large models have intensified, with commercialization, the choice between large and small models, and application directions emerging as key industry focal points.

So, how has the large model landscape evolved since 2023?

01. Philosophical Shift: From Idealism to Pragmatism

A notable change in 2024 is the market's increased pragmatism, prioritizing return on investment and commercial viability. A year ago, China's venture capital community buzzed with excitement over OpenAI and the promise of future AI advancements, drawing seasoned entrepreneurs like Li Kaifu, Wang Xiaochuan, and Wang Huiwen back into the fray. However, this fervor has waned, with investors growing impatient and conflicts arising between stakeholders. For instance, the prominent large model startup "Dark Side of the Moon" fell out with former investors, leading to arbitration that garnered widespread industry attention.

According to Juzi IT data, from January 1st to December 5th of this year, there were 439 financing cases in the domestic AI field, totaling over 56.4 billion yuan, roughly 80% of last year's total. The average monthly financing in the AI sector this year is estimated at less than 5 billion yuan, reflecting a decrease in institutional investment. According to Zero2IPO Research, early-stage investment, VC, and PE outbound activities in China declined by 23.9%, 19.2%, and 25.2% year-on-year in the first half of 2024, respectively.

If in 2023, large model vendors focused on refining model parameters, performance, and rankings, aiming to be "China's OpenAI," in 2024, the emphasis has shifted to practical applications and commercialization.

This trend intensified in the second half of the year, with the market's infatuation with large models waning, capital returning to rationality, and industry dynamics shifting. Giants like Alibaba and ByteDance accelerated investments, attracting talent from startups and launching AI products. Meanwhile, startups reevaluated their strategies, with some abandoning pre-trained models to concentrate on AI applications.

"Only 10 companies will significantly impact the AI large model sector," said Yaqin Zhang, a foreign academician of the Chinese Academy of Engineering. This consensus underscores that only a few players will ultimately prevail.

While the ultimate evolution of large models remains uncertain, technological advancements, captivating scenarios, and ongoing commercialization explorations persist amidst the large model wave.

02. Technological Roadmap: From Computational Might to Logical Learning

A second significant change in 2024 is the diversification of the technological roadmap, moving beyond mere computational power to explore reinforcement learning, knowledge computing, symbolic reasoning, neuromorphic computing, and other avenues. Smaller, more vertical models are emerging, complementing larger ones.

Large models have excelled in speech recognition, natural language processing, and image recognition, processing vast data volumes and extracting valuable insights through deep learning. They support various applications, particularly in finance, healthcare, and education, driving enterprise innovation. However, challenges persist, including high computational costs, domain-specific data requirements, and issues with interpretability and privacy.

Simply scaling up models brings challenges like inefficiency, hallucinations, and security risks. The core of current AI advancements lies in large language models (LLMs) with billions of parameters, requiring substantial GPU clusters and financial investments.

Open-source small models have emerged as a new frontier. While large models struggle with deployment on mobile and IoT devices, small models excel in efficiency, particularly in real-time applications like speech and image recognition. They offer computational efficiency and task-specific optimization, making them more interpretable and user-friendly.

03. Application Evolution: From Dialogue to AI Agents and Embodied Intelligence

The third notable change in 2024 is the continuous evolution of applications. With 467 large models globally, according to lifearchitect.ai data, vendors are iterating to approach AGI. Beyond diverse applications, a key shift is the limitations of large language models, exemplified by ChatGPT's constraints.

In response, OpenAI introduced the o1 series, emphasizing reasoning and problem-solving abilities. This sparked a domestic race, with vendors like Dark Side of the Moon, Deepseek, and Kunlun Wanwei unveiling models that emphasize logical thinking.

Meanwhile, "AI Agents" gained traction in 2024. Tech giants like Microsoft, Apple, Google, OpenAI, and Anthropic announced progress, while domestic platforms from Baidu, Alibaba, Tencent, and others entered the fray. An AI Agent, as defined by OpenAI, is an autonomous system that perceives, plans, and executes tasks, transcending traditional AI's limitations.

Additionally, embodied intelligence integrates AI into physical entities, enabling dynamic environmental interaction. Smart sweeping robots and self-driving cars exemplify this, with humanoid robots as the ideal platform.

These developments reflect the industry's quest for killer applications that maximize large models' potential.

04. Conclusion: Navigating Towards AGI

In summary, whether philosophical, roadmap, or application shifts, the ultimate goal remains AGI – Artificial General Intelligence. AGI aims to replicate human intelligence, encompassing learning, reasoning, perception, language, and creativity. Despite challenges, humanity's quest for knowledge and wisdom persists. With continuous advancements, AGI will likely become a reality, solving complex problems, enhancing quality of life, and potentially heralding a new era for human civilization. This journey must balance technological pursuit with ethical considerations, ensuring AGI's development fosters social equity and harmony.