Instead of delving into the potentially profitable realm of developing artificial intelligence-based software, the tech industry appears to be fixated on refining basic AI models.
Robin Li, the CEO of Baidu, suggests that major tech players engaged in creating the foundational models driving the current AI boom should shift their focus. Rather than continuously reworking these core models, he advocates for directing efforts towards crafting AI-driven applications that leverage existing models.
Li expressed his views at a business event in Shenzhen, emphasizing that the perpetual redevelopment of fundamental large models is a significant drain on resources. He stressed the need for a multitude of IoT-powered local applications to be developed at scale, numbering in the millions.
The complexity and cost associated with developing fundamental models, which encompass hundreds of millions, if not billions, of data points, pose challenges for all but the most advanced tech companies.
According to Li, while a large vocabulary model serves as a foundational element akin to an operating system, developers ultimately require fewer large models to create diverse local applications.
Despite attempts to seek input from Baidu, the company remained unresponsive.
Many leading tech companies in China and the United States have unveiled their guiding principles, with the issue of potential system ownership currently in the spotlight.
Baidu recently introduced Ernie 4.0, its latest AI model, positioning it as a competitor to OpenAI’s GPT-4. The unveiling of OpenAI’s ChatGPT AI-powered bot in November 2022 set a benchmark in AI models. In response to Meta’s Llama 2, a proponent of open-source AI technology, Baidu’s competitor Alfaba presented its own model. (Meta requires businesses with over 700 million monthly active users to obtain a special license to use its Llama 2 model.) Amazon took a slightly different approach by launching its Titan unit while also offering access to models developed by other creators. Li argued that competing with these tech giants, who are industry leaders, would be both impractical and futile.
Li envisions AI development following a trajectory similar to the smartphone revolution, where a proliferation of apps emerged following the widespread adoption of smartphones, leading to the rise of significant tech companies like Uber and TikTok’s counterpart, Douyin.
Li notes a lack of standout IoT-powered local applications in China and the US thus far, as companies and tech giants vie for market entry.
In China, as businesses strive to establish their presence, several competing AI foundational models have emerged, sparking what some refer to as a “war of a hundred versions.” Tencent, a Taiwanese company known for popular messaging apps like Snapchat, highlights concerns among experts that new entrants may struggle to compete against established players. Upon unveiling its AI model, Baidu experienced a 3% stock decline.
Li’s perspective contrasts with that of OpenAI CEO Sam Altman, who anticipates a shift towards smaller, specialized models tailored to specific use cases, supplanting the era of ever-expanding models.
The consolidation of these models within a select group of software companies, owing to the complexity and cost of development, raises questions about transparency in AI advancement. A Stanford University study revealed that the top ten AI developers in the US failed a “transparency test” assessing data sourcing, utilization, and model outcomes.
The Chinese government has taken steps to regulate the creation and deployment of large AI models, with the Cyberspace Administration of China issuing federal laws governing AI usage in July, potentially expanding the market for AI models in China.
