HackerNews中文版

我在美国一所R1公立大学攻读计算机工程博士学位。攻读博士学位是我一直以来的愿望，因为它能让我深入研究感兴趣的领域。但事实证明，最初的想法并不全面。我现在博士生涯已经接近第二年尾声。我的导师研究高性能计算，但并没有做前沿的研究。而我希望从事高效的硬件人工智能研究（更快的内核、机器学习系统等）。但每当我阅读论文，看到作者和他们的成就，以及他们背后的团队，还有那些进行研究的著名学校/实验室，而我却还在深入研究一些cuBLAS函数，希望能优化寄存器分块，以提升几纳秒的速度时，我就会觉得自己是个失败者。我不知道该研究什么问题。我甚至怀疑我想要做的事情是否真的需要一个博士学位。我质疑自己是否能做出任何可以发表的成果，那些MIT/斯坦福的实验室，以及他们拥有的那些杰出人才，几乎是单枪匹马就能完成研究？我还有太多东西不了解。我越觉得自己懂得多，就越有新的东西冒出来。我以为我终于理解了LLM的每一个细节，但现在又有更多新的注意力机制变体出现。我不确定自己是否能读完所有需要阅读的文献，从而能够进行研究。我希望在获得博士学位后能在美国找到工作（回国太可怕了），而且我还是国际学生，所以退学不是一个选项。我正在寻求任何建议，或者如果你也有类似的经历，任何建议都会有帮助。谢谢。

查看原文

I am doing a PhD in R1 public uni on Computer Engineering in the USA. PhD is something I really wanted to do because it allows me to dive deep into stuff. Turns out what I thought initially was not the whole story.I am nearing the end of second year of my PhD. My advisor is in HPC but doesn't do hot research and I want to do efficient hardware AI research (faster kernels, ML systems etc). But whenever I read papers and see the authors and what they have accomplished and the team behind them and the prestigious schools/labs doing them while I am deep in some cuBLAS function hoping to optimize register tiling to be a nanosecond faster, I feel like a failure. I don't know what problems to work on. I wonder if what I want to do even requires a PhD.I question if I will ever be able to do anything publishable as the labs in MIT/Stanford, their cohort of brilliant minds basically alone? There's so much out there that I don't know. The more I think I know the more stuffs keep popping up. I thought I had finally understood LLMs to their nitty gritty details but then there's even more new variants of attentions popping up. I am not sure if I will be able to read everything needed to be able to research. I want to be able to get a job in the US after my PhD(home is terrifying) and I am an international so quiting is not an option.I am looking for any advice or if you were in a similar boat anything would be helpful. Thank you.

问 HN：关于在读博士期间感觉自己很失败的建议？