我确实对事情的走向感觉不错。我们有很多来自 SpaceX 本部的员工,以前负责 Starlink 软件和 Starship 软件的人,现在在主导重写,主导新的 AI 软件栈,他们非常出色。
AI 训练有一个很有意思的地方:做训练时,你必须让大量 GPU 协同一致地工作,几十万块 GPU,而这会变成一个网络问题。就是如何让所有这些 GPU 以尽可能高效的方式互相通信?
所以这本质上是一个网络软件问题,而这恰恰也是 Starlink 在做的事——当你在太空中重建整个互联网时,这显然是一个网络问题。所以我们发现,这项技术和技能可以很好地迁移到 AI 训练上。
Um, and uh I do feel good about where things are going. Um we've got um uh a lot of the people from SpaceX proper um the the the you know we've got the the people that used to run Starlink software and uh Starship software now uh leading the uh rewrite you know leading the new um AI software stack um and and they're extremely good and and sort of an interesting thing about AI training is that when you do the training you you've got to have a large number of of GPUs work together coherently um hundreds of thousands of them and and it interestingly it becomes a network networking problem.
It's like how do you get all of these GPUs to talk together in in the most efficient manner possible? And this that so it's it's it's a yeah like a networking software problem essentially which is exactly what Star Link is also uh you know when you're rebuilding the entire internet and space uh it's it's a it's obviously a networking problem.
Um and so we found that that technology and skill is very transferable to uh AI training.