Hyperbolic的联合创始人兼CEO Jasper Zhang：
文字版
Just talked to the@deepseek_ai guys and here are some deep secrets:
刚刚跟 @deepseek_ai guys 聊了聊，这里有一些深藏的秘密：
V3 is just a start, they plan to release a new version in the next 3-6 months that are comparable to or even better than the latest GPT 4o model.
V3 只是开始，他们计划在接下来的 3-6 个月内发布一个新版本，该版本与最新的 GPT 4o 模型相当，甚至更好。
They are very research focused and never spent any dollars on marketing. The launch was not planned: it’s just that a few days ago the model reached a certain level so they decided to release it
他们非常注重研究，从未在营销上花过一分钱。发布并不是计划好的：只是几天前该模型达到了一定水平，所以他们决定发布
They believe in decentralization and democratization of AI models and will keep open sourcing new AI models
他们相信人工智能模型的去中心化和民主化，并将持续开源新的 AI 模型
Deepseek never received any VC funding. They came from a top hedge-fund called high-flyer (幻方).
Deepseek 从未获得任何风险投资。他们来自一家顶级对冲基金，名为幻方（high-flyer）。
A fun fact: Three years ago when I worked at Citadel, their cofounder wanted me to work with them (didn’t do it because I wanted to build my own startup). He told me that they built a data center for running ML experiments for predicting the markets and executing strategies but outside of the trading hours, most of the GPUs sit idle. Looks like they now find a good use of those idle GPU hours
一个有趣的事实：三年前我在 Citadel 工作时，他们的联合创始人想让我和他们一起工作（因为我想要建立自己的初创公司，所以没有这么做）。他告诉我，他们建立了一个数据中心来运行机器学习实验，预测市场并执行策略，但在交易时间之外，大部分 GPU 都处于闲置状态。看起来他们现在找到了利用这些闲置 GPU 时间的好方法
