Chinese Models have an advantage because Chinese symbols convey more meaning

This tweet from Jun Song (a Chinese/AI commentator) argues that Chinese AI models have a built-in advantage because Chinese characters (logograms) pack more semantic meaning per token than alphabetic scripts like English. In AI language models, "tokens" are the basic units the model processes — so if each Chinese character conveys as much information as multiple English letters or words, a model can achieve the same understanding with fewer tokens, potentially making it faster or more efficient. The context is the ongoing US-China AI competition, where Chinese models (like those from Baidu, Alibaba, or DeepSeek) often benchmark competitively despite using less computing power. This claim is debated: some researchers argue token efficiency is real, while others say any advantage is marginal or offset by hardware constraints.