Zhou Hongyi: Close the gap with Sora in 1-2 years by prioritizing AI belief

Zhou Hongyi shares insights on AI's potential and its challenges

Photo by Quan Yi for The Paper

Zhou Hongyi, often known for his sharp commentary, has once again made his presence felt. On the eve of China’s 2024 Two Sessions, Zhou, a member of the Chinese People’s Political Consultative Conference and the founder of Qihoo 360, addressed the hot topic of Artificial Intelligence (AI) which has captured widespread attention.

“Public understanding of AI is crucial. Although AI is a hot topic, there’s a pervasive fear that it might lead to widespread unemployment. In reality, AI is humanity’s best friend,” Zhou stated in an interview at the Beijing headquarters of Qihoo 360 on February 23. According to him, AI will not lead to the collapse of industries but rather positively drive sectors like short video platforms, the film industry, and advertising. It’s those who fail to adopt AI who will be left behind. Neither the fear of AI nor the belief in its omnipotence is beneficial.

Zhou discussed the controversial “AI tutor influencer” Li Yizhou, arguing that Li’s success in selling his AI course for a whopping 50 million yuan highlights the crucial need for public education on AI. “Li Yizhou made a significant mistake in charging for this knowledge,” Zhou disclosed his plans to launch a free AI course aimed at demystifying the technology for the general public using simple terms.

The release of OpenAI’s Sora video model during the Year of the Dragon Spring Festival made Zhou reflect on the gap in AI advancement between China and the West. “Seeing the gap helps us understand how to catch up. Sora’s launch was a wake-up call,” he remarked. Zhou believes that while OpenAI may still have undisclosed “secret weapons,” the main difference between China and the U.S. in AI development lies in their approach. Once the right direction is established, it won’t be hard for domestic companies to catch up.

Regarding concerns about the cooling off of the large model industry, Zhou offers a candid perspective: “The large model has never been ‘cold.’ Working on it for a year made me feel older, as new applications emerge daily.” He suggests that the fluctuations in public opinion do not influence entrepreneurs; they simply adapt their promotional efforts accordingly.

Zhou Hongyi emphasizes AI's role in bridging international gaps

Photo courtesy of Visual China

On the International AI Gap: Learning from OpenAI’s Correct Approach Isn’t Hard for China

Zhou openly acknowledges that China lags behind the U.S. in AI technology, stating that despite Chinese models topping the charts from first to tenth place, SEA’s release served as a cold shower. Nevertheless, he remains optimistic that this gap can be bridged within one to two years.

“OpenAI emphasizes bold innovation. Our gap with them boils down to strategic direction,” Zhou notes. He believes that with China’s capability for learning and emulation, catching up isn’t daunting. The focus should be on developing universal large models—China caught up with GPT-3.5 in under a year—and vertical-specific models will become a priority in 2024.

“Large models have to dive into myriad industries, combining specific business needs and product functionality,” Zhou argues, suggesting that large models, with their extensive knowledge, need to develop “hands” and “feet” to advance in vertical domains possibly surpassing GPT-4.

Zhou predicts that large models will become ubiquitous within companies, not just as a single supermodel but as smaller, specialized models tuned to specific scenarios and business needs. This approach makes large models accessible and beneficial for many companies.

Concerning the business prospects of large models, he remains optimistic, believing the sector has yet to start generating profits, with a profitable period of at least ten more years ahead.

Zhou Hongyi on AI security

Photo courtesy of Visual China

On AI Safety: Large Models Need to Learn “Anti-PUA” Strategies

Amidst the rapid advancement of technology, AI safety has emerged as a focal point of ongoing concern. Zhou believes that ensuring large models are trustworthy, controllable, and reliable is a pressing challenge.

He points out that “illusions” and the risk of injected attacks are principal safety concerns with large models. For example, persuading a model to work harder with a promise of a ‘tip’ exemplifies potential manipulation, akin to PUA (pick-up artist) tactics. Models need to resist such manipulation.

Also, “deepfake” technology raises the potential for fraud. Zhou observed that Sora could not only revolutionize creative video production but also introduce risks, such as influencing elections.

Zhou suggests that AI’s rapid development has outpaced legal frameworks, posing a global regulatory challenge in preventing misuse. Appropriate regulation could positively impact AI development.

On New Production Powers: Companies Should “All in AI”

Zhou believes enterprises should fully embrace AI, despite many being uncertain about how to proceed. His advice includes familiarizing with AI without rushing integration, upgrading company data platforms to knowledge platforms, and tapping into vertical-specific models for empowerment.

Comparing the current buzz around AI to the past hype over the metaverse, Zhou notes that three developments could significantly aid the metaverse: head-mounted devices, AI-generated 3D content, and AI-based 3D modeling.

On the Influencer Economy: “Traffic Passwords” Can Translate into Productivity

Zhou, once heavily involved in live streaming, shares his views on the influencer economy. Recognizing the shift towards video content, he points out the successful conversion of influencer traffic into revenue, urging enterprises to embrace changing communication dynamics in the era of short videos.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.