Tang Xiaoou and His Unfinished AI Puzzle

By Lili Yu

After September 2021, Tang Xiao'ou stopped updating his WeChat Moments. Now, he never will.

On December 16, news of Tang's death began circulating online. Shortly after, SenseTime posted an obituary on its official account: Tang Xiao'ou, founder of SenseTime, AI scientist, director of the Shanghai Artificial Intelligence Laboratory and the Pujiang Laboratory, and professor at the Chinese University of Hong Kong (CUHK), had passed away on December 15, 2023, after unsuccessful medical treatment.

Tang was widely regarded as a pioneer of computer vision in China and "a global trailblazer in facial recognition technology." The Multimedia Laboratory he founded at CUHK became known as the Whampoa Military Academy of China's AI vision research.

The lab's original team would later form the precursor to SenseTime. Cofounders Xu Li and Xiaogang Wang both came through this lab.

At its peak, SenseTime was the most highly funded and most valuable unicorn in AI globally. When it listed in Hong Kong in late December 2021, its market cap briefly exceeded HK$150 billion. At the time, it was the undisputed leader in China's computer vision (CV) track and among the "AI Four Dragons."

But history takes unexpected turns. ChatGPT's explosive arrival in 2023 instantly rendered the CV approach a narrative of a bygone era. In the second half of 2022, as SenseTime's revenue and growth ceiling lost their luster, its stock began a steady decline. In late November, US short-seller Grizzly Research even published a report accusing SenseTime of artificially inflating revenue through highly suspicious circular transactions, sending the stock to a new yearly low.

This April, SenseTime released its own large language model, which received regulatory approval for launch in August. No one knows what becomes of SenseTime in the LLM era. Nor will anyone know what psychological struggles Tang endured as one of the earliest scientists to cross from the lab into industry.

All of this is now beyond Tang's concern.

An AI scholar close to Tang told us that the name "SenseTime" (商汤, Shang Tang) draws from a period of Chinese history marked by transformative creation. "During the Shang Tang era, agriculture and handicrafts developed rapidly; oracle bone script emerged and found practical use," the scholar said. "It was a crucial stage in the progress of ancient civilization." SenseTime undoubtedly harbored similar ambitions.

From the company's name alone, one can see the story of a 55-year-old scientist whose grand ambitions remain unfulfilled.

Crossing the River by Feeling the Stones

For Tang, 2014 was a decisive year.

That year, his lab team at CUHK published a facial recognition algorithm that, for the first time globally, surpassed human-level recognition accuracy — something previously considered insurmountable.

Perhaps sensing this, Niu Kuiguang of IDG Capital caught the scent of commercial potential in vision technology and rushed to Hong Kong to meet Tang. IDG's bet on SenseTime at the angel round made it the biggest institutional winner at the company's IPO, while Tang himself, through this investment, began his transition from reclusive researcher to businessman. SenseTime was formally established that same year. Many joked at the time that the name "SenseTime" (商汤) was really a pun on Tang going into business (经商的汤).

CEO Xu Li has recounted on many occasions how he joined SenseTime. Tang felt that what he was doing then wasn't "powerful enough," and that "building SenseTime had a purpose: to build influence, so we could do bigger and greater things." Xu saw it as a long-term vision to change the world, and signed on.

Before this, Tang had followed a purely academic path.

Born in 1968 in Anshan, Liaoning Province, Tang was gifted from an early age. After graduating from the University of Science and Technology of China (USTC) in 1990, he went to the US for further study. Following his PhD from MIT, he joined CUHK and founded the university's Multimedia Laboratory from scratch.

Like many professors freshly emerged from the ivory tower, Tang initially learned entrepreneurship as he went.

During a summer 2014 visit to Silicon Valley, he didn't arrange meetings with heavyweight figures. Instead, he scheduled time with Facebook researchers and engineers. Through these conversations, he tried to understand what a startup should look like — down to granular details like office layout and kitchen design.

Between 2016 and 2018, SenseTime completed eight consecutive funding rounds. The frenzy of over thirty investors pouring capital allowed the company to rapidly scale from dozens to thousands of employees.

SenseTime's talent density once amazed many tech companies. Data from June 2021 showed the company had 40 professors, over 250 PhDs and doctoral candidates, and 3,593 scientists and engineers on its R&D team.

Regarding Tang's management style, a longtime SenseTime employee told us that internally, Tang always maintained an "understanding, gentle" attitude toward young people and students. Yan Shuicheng — now co-CEO of Kunlun Tech's Tiangong AI and dean of the 2050 Global Research Institute — was the first Tsinghua or Peking University graduate to join Tang's MMLab. In his view, Professor Tang was "especially calm, patient, and guiding," exceptionally skilled at "helping students amplify their capabilities."

One of Tang's best-known management philosophies was the "Black Sheep" culture. After observing the wolf-like cultures of many large companies, he explicitly rejected them, believing SenseTime should become an empathetic tech company. So they would be sheep, but not sheep with herd mentality — rather, "black sheep that do what others haven't done."

SenseTime on a Knife's Edge

Returning to 2014: when Tang crossed into business, what he likely didn't anticipate was that he would from then on be wrestling with far greater uncertainty.

Starting from computer vision, SenseTime's AI reach gradually extended into smart business, smart cities, smart living, and smart vehicles.

Yet even by 2021, when Tang's net worth reached 53.4 billion yuan at the IPO, many still questioned the company's long-term cash-burning finances, its uncertain business model, and its scientist-dominated governance structure. Later-stage departures were notably visible.

In previous interviews, Song Chunyu, VP at Lenovo Group and partner at Lenovo Capital, attributed this mainly to insufficient industry confidence in AI. The "AI Four Dragons" had burned through capital for so long without establishing a virtuous commercial cycle, he argued, so industry began to worry. "AI is highly complex technology; its deployment requires a long chain from data to training to application."

In fact, the much-criticized SenseTime also serves as a microcosm of Chinese AI's commercial logic and development trajectory over the past decade.

Song noted that companies like SenseTime and Megvii missed the opportunity to create the latest generative AI because they were largely forced by reality into systems integration work.

AI isn't like consumer internet, where business model replication and rapid capital influx can quickly build scale. So the previous wave of Chinese AI companies mostly landed on B2B commercialization models. The essence of this model is customized "outsourcing," and each scenario's uniqueness drives massive customization costs.

Moreover, when entering vertical tracks, they face competition from traditional industry leaders adding AI; when building platforms, they get squeezed by BAT internet giants.

SenseTime, perpetually searching for survival space between these two forces, was undoubtedly anxious — not to mention AI companies' towering personnel costs, hardware investment, and compute expenses.

How should SenseTime's puzzle be solved?

Song believes: "Chinese technology needs sufficient tolerance for exploratory breakthroughs in fundamental research." "AI's business model is hard to find, but it's a super-complex technological project for humanity, equivalent to the Manhattan Project. If it breaks through, it represents the greatest advance for all humankind." He sometimes struggles to understand why bottled water companies can command hundred-billion valuations while some high-tech companies, at just tens of billions, are harshly demanded to turn a profit.

However it might be solved, Tang Xiao'ou will not be the one solving it.

This scientist, who disliked media exposure and rarely gave interviews, was in the eyes of those around him a devoted parent who loved showing off his children, a film buff, and a "master comedian delayed by artificial intelligence" who perfectly inherited the Northeastern Chinese gift for humor.

His most recent public appearance on record was this July in Shanghai.

At the opening ceremony of the 2023 World Artificial Intelligence Conference (WAIC), Tang shared the stories of three students — Xiaogang Wang, Kaiming He, and Dahua Lin — and their pursuit of dreams in deep learning.

At the end of his speech, he once again thanked the students and teachers he had worked with, quoting a line from Yu Qian in the film Song of Youth: "I didn't meet you in the best of times; meeting you made this the best of times."

Several longtime SenseTime employees told us that at the company, Tang was accustomed to being called "Professor Tang." And on his WeChat profile, which will never update again, the identity he still listed was: "Tang Xiao'ou, Chinese University of Hong Kong."

Image source: IC photo

Layout: Yunxiao Guo