Research

MLSys25 talk from lidong: the next horizon of systems

MLSys25 talk from lidong: the next horizon of systems

image.png

image.png

以往的opti,尤其是对着profiling去分析其实是很耗费时间的,哪怕一个专业的研究员也要2天。

image.png

image.png

The grand opportunity is to use AI to confront technical debt in systems at its root, creating new methodologies that make large-scale systems more tractable, trustworthy, and evolvable.

What are systems as a discipline in the first place? Systems is the discipline of turning complexity into order: discovering the abstractions, interfaces, and invariants that make computing systems understandable, manageable, evolvable, and trustworthy

image.png

我们能用AI做什么?一个是适配软硬件,一个是更好的系统结构,还有就是控制抽象。 image.png

image.png

image.png

但是用AI来做形式化验证,还是会存在一些问题,AI为了追求pass而走了错误的shortcut。比如跳过/简化/转移一些困难的代码证明点。

image.png

image.png

image.png

一个verification native 的编程语言。

image.png

If you need to cite this post

Haibin Lai. (October 25, 2026). “MLSys25 talk from lidong: the next horizon of systems” [Blog post]. Retrieved from https://www.haibinlaiblog.top/posts/mlsys25-talk-from-lidong-the-next-horizon-of-systems/

@online{lai-mlsys25-talk-from-lidong-the-next-horizon-of-systems,
  title={MLSys25 talk from lidong: the next horizon of systems},
  author={Haibin Lai},
  year={2026},
  url={https://www.haibinlaiblog.top/posts/mlsys25-talk-from-lidong-the-next-horizon-of-systems/},
}