On DeepSeek and Export Controls

China is now second only to the United States in AI investment. But the quality of actual AI research, especially in the field of generative AI, has been hindered by government censorship and a lack of indigenous intellectual property. In fact, many of the Chinese AI startups that have taken advantage of the strong government support are producing

... See more

转：关于 DeepSeek 的研究和思考 (Archerman Capital) 关于这几天很火的 DeepSeek, 我们 (Archerman Capital) 做了一些研究和思考, 和大家分享, enjoy! 灰色部分是技术细节, 不感兴趣的可略过。 几个事实 1) DeepSeek 不是套壳不是蒸馏美国的大模型。 虽然中国有些大模型是套壳和蒸馏的, 但 DeepSeek 不是。 2) 核心架构还是基于 Transformer, deepseek 在架构、工程设计上进行了创新和工艺提升, 实现效率优化。架构上, 采用了混合专家模型 (MoE)、多头潜注意力 (MLA)、多令牌预测 (MTP)、长链式推理 (CoT)、DualPipe 算法等设计, 并进行了依赖强化学习 (RL)... See more

what OpenAI, Anthropic, DeepMind have all tried to do is raise billions & tap vast GPU resources of tech giants without having the resulting tech de facto controlled by them. I'm arguing the OpenAI fracas show that might be impossible.

LLMs are a moat, but for how long?

An LLM vendor like OpenAI isn’t an aggregator, As far as I can tell.

• An aggregator leverages a monopoly on demand to commodify supply .

• Whereas a traditional industrial monopoly leverages a monopoly on supply to extract $ from demand .

LLM vendors seem more like a traditional industrial monopoly. Like an industr

... See more