AI Hype Train Stalled in China

AI Hype Train Stalled in China

Chinese Startups Navigate the World of Artificial Intelligence in an Era of “Involution”

AI Startups

In the bustling tech landscape of China, startup founders like Zhang Haiwei face a daunting challenge – building their own large language model (LLM) seems like an impossible feat. The cost, competition, and lack of computing power make it nearly impossible for newcomers to break out. However, instead of wallowing in despair, Zhang’s motion-capture startup, Chingmu, has found an alternative approach. They leverage OpenAI’s pre-trained models to analyze human and object movement for animation and sports training.

This phenomenon, known as “involution,” has become increasingly prevalent in China’s tech sector. It describes the cycle of intense competition where individuals work harder for diminishing rewards. Chinese startups, like their counterparts in Silicon Valley, initially rode the wave of excitement around artificial intelligence. However, as the dust settled, it became clear that revolutionary applications were still elusive. Limited supplies of graphics processing units (GPUs), crucial for AI development, added another layer of challenge to the already complex situation.

Now, AI has become more about making small efficiency improvements and gaining a competitive edge in a challenging economic environment. With pressure mounting, startups and big tech companies are focusing on justifying the initial hype and excitement by delivering tangible results. Kevin Xu, a tech investor and founder of the AI newsletter Interconnected, notes this shift in focus.

However, despite the official blocking of OpenAI’s uncanny chatbot ChatGPT in China, it became the talk of the town. Internet users scaled the Great Firewall to try it out, creating a black market for the service. The frenzy surrounding ChatGPT sparked renewed interest in AI and revitalized a sector that appeared stagnant. Some large companies even regretted firing their teams working on large language models when the technology seemed less promising.

One major hurdle Chinese startups face in developing AI models is the scarcity of data. Unlike their American counterparts, who had access to vast amounts of data from sources like Google and social media platforms, Chinese companies faced significant challenges. China skipped the open web and directly embraced apps, which are much harder to scrape for data. Coupled with the high cost of computing power, this obstacle obstructs the creation of the sweeping models seen in the United States. As a result, many Chinese startups are focusing on the application level rather than building their own models.

Vrch.io, an AI startup founded last year, exemplifies this trend. The company is developing an AI-powered voice-entry image generator, allowing interior designers to easily communicate their ideas to clients. By leveraging generative AI, the time-consuming process of converting information into images to ensure accurate representation is streamlined. Vrch.io currently targets overseas markets due to regulatory uncertainties in China. While interim guidelines on generative AI have been released, specific details are still awaited. However, it is apparent that Chinese regulators aim to strike a balance between encouraging innovation and maintaining control over technology development.

Despite the economic challenges facing China, the tech sector continues to adapt and innovate. Startups like Lingua Technologies are using AI to address specific pain points. Lingua Technologies aims to compete with translation companies in Beijing, providing a solution to the cognitive strain placed on native English speakers working with Chinese academics. By leveraging AI for tedious tasks such as formatting, Lingua Technologies reduces the costs associated with knowledge work, making cross-cultural collaboration smoother and more efficient.

Further exemplifying the potential of AI, Medlinker, a Chinese-focused healthcare startup, held a livestreamed contest between its AI doctor, “MedGPT,” and real doctors. Although MedGPT asked more questions than a human doctor typically would, it successfully reached the same diagnosis. MedGPT serves as a channel to connect patients with Medlinker’s existing infrastructure of hospitals, demonstrating the potential of AI integration in the healthcare sector.

Venture capitalists in China are cautiously navigating the evolving AI landscape. They are more discerning in their investments, recognizing the potential downsides in the current macroeconomic climate. However, this lull in investment activity is not indicative of a shift away from AI. Considering the nature of AI’s long development cycle, VCs are merely waiting to see the competition unfold before making their moves.

In conclusion, Chinese startups are navigating the world of artificial intelligence amidst a paradigm of involution. While the initial hype has subsided, companies are leveraging AI for small efficiency gains and staying competitive. The scarcity of data and computing power remains a challenge, prompting startups to focus on applications rather than building their own models. Despite the economic uncertainties, the tech sector continues to find innovative applications for AI. As China’s regulations on AI take shape, startups are preparing to seize the opportunity to make their mark on the global AI stage.