The US wants to keep GenAI out of China, but that’s an impossible task

The US government is working on plans to keep AI software out of China’s hands. The Commerce Department is considering banning the export of closed-source AI models. This means that it does not contain open source options that cannot be controlled in any way.

with plans, revealed Reuters, sources talk about “guards” keeping advanced AI models at bay. This is about the limited distribution of LLMs like OpenAI’s GPT-4 or Google Gemini, not their functionality. Currently, AI vendors can sell their technology to anyone, including the Chinese government, without any restrictions.

The Biden administration fears using these models to wage cyberwarfare. In that sense, the step seems to be mainly prevention. Future LLMs may give new impetus to cyber attacks, but currently the use of GenAI technology is limited to writing more convincing phishing emails than before.

Llama 3 problem

Central to the project are parties like OpenAI, Google and Anthropic. These companies are leading AI players and house their most advanced models. LLMs can only be used through their own platforms or APIs; For example, it is impossible to run GPT-4 locally.

This does not apply to Meta’s Llama models, version 3 of which appeared recently. Since Llama 1 opened early last year, these open source LLMs have been considered State of the art For free use of GenAI. Any developer can start with a variety of llama models that can be trained, fine-tuned, or basically adjusted as desired. If the definitions are to be believed, this does not mean that LLMs are significantly inferior to the offerings of OpenAI/Google/Anthropic.

read more: Meta releases powerful open source model Llama 3 and chatbot Meta AI

Where is the border?

The question is how any US projects will deal with this open source problem. Various IT companies have long criticized the potential impact of regulations on open source. For example, the Cyber ​​Regression Act threatened to limit contributions to open source because developers would suddenly be responsible for code someone else was using unsolicited for their own purposes. That danger seems to have passed now.

Either way, the Biden administration wants to put AI technology to the test. AI models can be scored for intelligence via all sorts of benchmarks, so a certain score can determine whether an LLM can be shared with China. This is reminiscent of current US legislation that prohibits the export of AI hardware above a certain level and imposes restrictions on ASML machines.

These earlier export controls appear to have been partially successful. For example, Chinese parties still produce significantly more advanced chips with older machines, although their precise quality is questionable. The latest ASML machines cannot be launched without the expertise of the Veldhoven company. It also doesn’t apply to Nvidia’s GPUs running AI models like GPT-4 and Gemini. China has access to AI chips that should not be under restrictions, it has recently emerged. It cannot be ruled out that faster AI hardware will also be sourced from third parties in the US, Europe or elsewhere.

How successful America’s technology war against China is once again in the shadows. Open-source AI development can be a thorn in Washington’s desire to protect AI from Beijing. China itself has been trying to reduce its dependence on the West for some time. ASML’s former CEO, Peter Wennink, has already predicted this move: Blocking exports to China will force the country to advance technologically.

read more: China bans Windows, Intel and AMD from government use