China's artificial intelligence strategy scares the US. The case of Qwen 2.5

It is no mystery, in fact, that the main producers of generative artificial intelligence are all stars and stripes Moreover, Washington's blockade of Beijing over Nvidia's chips, which to date are the most advanced for learning generative AI algorithms, represents a bottleneck to China's progress in this field. To put it another way, the American advantage is as of today out of the question.

Qwen's success, on the other hand, seems partly related to its size. In fact, we are talking about a 1.5 billion parameter model. ChatGPT 3.5, the chatbot that debuted in 2022, had 175 billion, and the current one, GPT-4o, is even bigger. We are therefore talking, in the case of Qwen, about a very small model that can be downloaded even on the cheapest smartphones of 2024. It does not need to connect to external servers, but can be run locally, so it does not share information outside the device. In essence, it does what ChatGPT does, but in a much more limited way: it answers questions with natural language.

It supports over 29 languages, including Italian, English, French, Spanish, German, Russian, Japanese and Korean, and can manage up to 32,768 tokens and generate texts of up to 8,192 tokens. In simple terms, tokens correspond to pieces of words, where 1,000 tokens are roughly equivalent to 750 words.

Technologically, it is not something new.

Qwen is successful precisely because, as many people on Reddit suggest, especially in third world countries, they no longer own a computer, but only a smartphone, usually with little memory. And these small, open systems are perfect. On the ball, of course, are not only the Chinese but also Big Tech itself, which have added smaller, open and free versions to their proprietary models. As in the case of Google, which in addition to Gemini has Gemma.

So it is not necessarily the case that open source is the focus of China's strategy. On the contrary, it is by no means certain that in the short term the goal is to compete on an equal footing with US Big Tech. Trump can rest easy but up to a point because the challenge will not be a flagship one. We are not witnessing a race but a marathon. Technological development in the Red Continent, as anyone who has studied this country closely (and for some time) knows, never proceeds at a breakneck pace. And it is never what you expect. There has been concern in recent weeks about the news that the Chinese army has started using an adapted version of the American Meta's open source artificial intelligence Llama for military purposes. Those expecting a compact and flagship advance might be disappointed. China in Ai seems increasingly ecumenical and agnostic. To be near and close, but not in the way you expect.