The Biden administration is reportedly working on protecting US technology from China and Russia, with a particular focus on limiting the export of advanced AI technologies, including OpenAI’s ChatGPT and closed-source AI systems trained on undisclosed data, according to a recent report from Reuters.
This raises the question of how advanced Chinese AI technologies currently are and whether they rely on US tech to develop their own domestic AI models. Let’s delve into this further.
When OpenAI released ChatGPT at the end of 2022, it surprised many, including China. However, in 2023, Chinese companies started releasing AI generative chatbots after receiving approval from their government. Since OpenAI is not available in mainland China, how did this happen?
Many Chinese companies and engineers gained access to ChatGPT through proxy servers and VPNs, enabling them to develop their own applications and software using OpenAI’s models. Most Chinese companies benchmark their models against OpenAI’s models.
In February, OpenAI took action against two China-affiliated threat actors, known as Charcoal Typhoon and Salmon Typhoon, preventing them from accessing ChatGPT. These companies had been using OpenAI’s services for tasks such as translating technical papers, debugging code, and generating scripts. In December, OpenAI also suspended accounts belonging to the Chinese giant ByteDance for using ChatGPT to create a rival AI model.
China is currently playing catch-up in the field of artificial intelligence. There is a significant gap between OpenAI and Chinese AI technology. While some Chinese companies have achieved success in building AI models, they have not reached the level of ChatGPT. This is mainly due to the lack of access to data, as most available data sources are in English.
Building AI models poses challenges for China, as there are restrictions on accessing closed-source AI models and advanced GPUs. In 2022, Nvidia and AMD banned China from purchasing their most powerful AI chips. However, this has not stopped the country from finding alternatives.
In terms of software, China is relying on open-source AI models like Meta’s Llama LLM models. Wired reports that a startup called 01.AI gained significant success after releasing Yi-34B on HuggingFaces. Yi-34B outperformed Llama 2 and became one of the recommended models for developers. However, 01.AI faced backlash when developers discovered that Yi-34B was based on Llama.
Regarding hardware, Huawei is leading a group of semiconductor companies in manufacturing local AI chips. The Chinese company is competing with Nvidia and striving to autonomously develop advanced AI chips. Huawei is working on designing and manufacturing high-bandwidth memory chips from scratch, which are crucial components of advanced AI GPUs.
Despite these efforts, China still lags behind in the AI race and has a long road ahead. OpenAI has a significant advantage due to its early start, as creating an AI model involves training algorithms and accessing a vast data pool. China currently falls short in both areas and will continue to depend on Western technology, at least for the time being.
