I’d guess by the end of 2025, it’s almost guaranteed locally run Chinese AI models will become by far the most common and used forms of AI around the world.
So far the Chinese AIs have been using the global corpus of internet scraped data, but they are about to get a new source - Chinese public data. This raises an interesting possibility. First, that Chinese AI starts thinking in a more Chinese way, and second that this form of thinking may become the most dominant form of AI thinking globally.
There’s nothing stopping Western companies from using the same method to make more efficient AI.
The profit motive does. If AI gets efficient enough, you don’t need them to host it. Deepseek can run on a high end desktop GPU already.
So can llama and gemma