Sublime

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

Economists have long day-dreamed about problems that could be solved by market mechanisms. The laboratory has been opened, and by simply issuing a coin and a few rules, you are free to run your own cryptoeconomic experiment.

I see an experiment with care as an untapped resource with the potential to undo capitalism.

Networks are experimenting with lockup mechanisms for ongoing emissions in an attempt to improve price volatility and exclude speculators. 

There are still new science startups being built today, such as New Limit, Arcadia Science, and Altos Labs. But now there are also examples of research institutes, such as Arc Institute and New Science, and there are even emerging examples of crypto-native experiments, like VitaDAO and LabDAO. It’s not that one approach replaced the other, but that... See more

Building something of significant value to the student, but shifting the risk away from the student to your company or the current or future employer is the way to build a very large education company. Prenda does this for parents in K12 by serving as a public charter school, Lambda does this for students by deferring tuition until they get a job.