The AI war between US and China heats up as Alibaba open-sources two models

Both Qwen-7B and Qwen-7B-Chat are Large Language Models (LLM), which is a type of Artificial Intelligence (AI) model designed to understand and interact in natural human language. (Image: Reuters)

Chinese technology giant Alibaba's cloud computing unit, has open-sourced two of its Artificial Intelligence (AI) models to take the battle to Meta's Llama 2.

Both Qwen-7B and Qwen-7B-Chat are Large Language Models (LLM), which is a type of AI designed to understand and interact in natural human language.

These models such as ChatGPT, Bard or Bing AI, can generate text, answer questions, translate languages, and a lot more. The reason they are known as LLMs is because they are trained on massive amounts of data, and have billions of parameters to work with.

Qwen-7B and Qwen-7B-Chat are smaller versions of Alibaba's largest language model, Tongyi Qianwen, which can be roughly translated to "seeking truth by asking a thousand questions".

Alibaba first announced the model in April when the company's CTO Zhou Jingren said that they will allow enterprises to build services by opening up the functions of the model.

The AI model made headlines when it was announced that would have 10 trillion parameters. By comparison, OpenAI's GPT-4, which underpins ChatGPT, has 1.7 trillion parameters. Qwen-7B and 7B-Chat have 7 billion parameters in total each.

The company plans to integrate the capabilities of the model into its existing platforms such as DingTalk, an enterprise communication platform; and Tmall Genie, which is a smart speaker that uses Alibaba's virtual assistant software, AliGenie.

Unlike Qwen, Meta's Llama 2 is available in three sizes - 7 billion parameters, 13 billion parameters and 70 billion parameters, allowing applications from small businesses to large scale enterprises. In comparison, Qwen is aimed small and medium size businesses who want to overhaul their workflow with AI.

Similar to Meta, Alibaba will make the code, weights and documentation available to the public for commercial use, but there is a catch.

Organizations that run services which have over 100 million monthly active users, will have to pay for a license fee to use the model. Similarly, Meta's cut-off limit is 700 million monthly active users.

If you look at the current AI landscape, most of the power lies with big corporations such as Microsoft, Meta, Google, and OpenAI, etc.

Open-sourcing language models allows other companies to compete in this space. While this may not mean they will overtake the current darlings of AI, they will create competition and help democratise the market, giving more options to potential users.

Another big reason is mini-war over AI that the United States and China are engaging in. China wants desperately to break into the AI market, and the local government encourages players such as Alibaba, Tencent and Huawei to develop competitive models.

AI is a rapidly evolving, highly competitive field which may decide future power stakes for countries. The US is also tightening the screws on exports of AI-chips to China, and even taking steps to restrict cloud-computing access to its AI rival.