I. AI Is Inevitable, But Crypto x AI? At the beginning of 2025, DeepSeek, created by the quantitative firm Huanfang, dropped a "nuclear bomb" in the AI community. A Chinese AI model that used only 2,048 NVIDIA H800 GPUs and had a training cost of $5.58 million (about one-tenth of Meta's) achieved results comparable to GPT-4o and Llama 3.1 in benchmark tests like MMLU and GPQA. It even slightly surpassed these top Silicon Valley models in complex reasoning and Chinese semantic understanding. D...

I. AI Is Inevitable, But Crypto x AI? At the beginning of 2025, DeepSeek, created by the quantitative firm Huanfang, dropped a "nuclear bomb" in the AI community. A Chinese AI model that used only 2,048 NVIDIA H800 GPUs and had a training cost of $5.58 million (about one-tenth of Meta's) achieved results comparable to GPT-4o and Llama 3.1 in benchmark tests like MMLU and GPQA. It even slightly surpassed these top Silicon Valley models in complex reasoning and Chinese semantic understanding. D...