Chinese tech heavyweight SenseTime recently released SenseNova 5.5 at the World Artificial Intelligence Conference in Shanghai, claiming a 30% performance increase over its predecessor and superiority over GPT-4o in several criteria.
The company said that key enhancements include improved mathematical reasoning, English proficiency, and command following capabilities, putting it on par with GPT-4o in terms of interactivity and other core indicators.
In early May, SenseTime released a demo similar to OpenAI’s GPT-4o demo, showing off the model’s visual skills. SenseNova 5.5 can recognise and describe specific items by pointing a smartphone camera at them while the AI operates.
Similarly, Baidu USA CEO Robin Li claimed that Ernie 4.0 can produce better results than GPT-4o. Meanwhile, Alibaba’s Tongyi Qianwen (Qwen) models saw downloads shooting up to 20 million, tripling in only two months.
Forget text generation—China has leaped ahead in video generation models too. While the world awaits the official release of Sora, the internet is abuzz with Kling’s videos depicting a series of animals enjoying a meal of noodles.
In the meantime, OpenAI recently announced that as of July 9, it had blocked API access for users in unsupported countries and territories, including mainland China, Hong Kong, and Macau.
China’s Race To Dominance
Hugging Face co-founder and CEO Clement Delangue praised the progress made by Chinese AI firms in a post on X, saying, ‘Qwen 72B is the king, and Chinese open models are dominating overall.'” This is proven by the fact that Alibaba’s Qwen 2-72B model has claimed the top spot on Hugging Face’s current LLM Leaderboard, outperforming all other open-source models.
“China’s advantage is doing whatever it takes to catch up,” said Kai-Fu Lee, a Taiwanese computer scientist and founder of China-based AI startup 01.AI. Lee’s firm open-sourced Yi-34B, its foundational LLM that outperformed Llama 2 on various benchmarks.
He went on to explain how 14 months ago, they had nothing, including no GPT. “At that time, we were six or seven years behind, and at this moment, we are six to nine months behind. So the catch-up has already been happening. Going forward, we hope that it will continue,” Lee said.
In May, China-based DeepSeek open-sourced its DeepSeek LLM, a 67-billion parameter model trained from scratch on a dataset consisting of 2 trillion tokens in both English and Chinese, hinting at its bid to go global. The model managed to outperform Llama 2, Claude 2, and Grok-1 in various metrics.
Government Support
Chinese tech companies have also been receiving support from the government in its ongoing AI battle against America. The Chinese government has made AI a national priority, aiming to become the world leader in AI by 2030.
The Cyberspace Administration of China (CAC) has issued approvals for over 40 LLMs in the past six months, granting operational licences to 1,432 AI-driven applications.
Meanwhile, according to a survey, over 80% of Chinese business leaders surveyed are currently using GenAI in their operations, way above the global average of 54% and the US average of 65%.
China also dominates the global race in GenAI patents, filing more than 38,000 patents from 2014 to 2023, according to a UN report. That’s six times more than those filed by the US-based inventors. Geographically, China leads with 38,210 inventions, far surpassing the US (6,276), South Korea (4,155), Japan (3,409) and India (1,350).
With the release of these models, the fear of China rising against US open-source models, particularly in the field of generative AI, is not unfounded. This competition is driving significant advancements in AI technology.
Just a Copycat?
But despite the nation’s mad rush to develop generative AI, Chinese businesses are almost wholly dependent on American underpinnings, such as open-sourced foundational research and technology developed by leading US companies and research institutions.
“As a measure of how far behind they are, leading Chinese firms are comparing their performance to ChatGPT,” said Paul Triolo, technology policy lead and senior VP for China, Dentons Global Advisors.
China’s businesses typically use “fine-tuned versions of Western models” because their own AI models “aren’t very good,” said Jenny Xiao, a partner at San Franciscan venture capital firm Leonis Capital. She added that Silicon Valley is unquestionably far ahead of the curve.
For instance, some of the technology in Chinese firm 01.AI, which released its open-source model, came from LLaMA. Former Google CEO Eric Schmidt said that while China intends to take the lead in several industries, the US is still far ahead in artificial intelligence.
