1d ago

Databricks Mosaic AI's Andrew Drozdov argues computer science is within two years of being too complex for one expert to master

Boytsov notes GPU specialization already exceeds standard curricula.

0
Original post

There was a point where a mathematician could no longer understand all math. I suspect we’re close to (1-2 years) an equivalent inflection point for CS, in particular wrt AI infrastructure.

12:06 AM · May 26, 2026 View on X

@mrdrozdov You took a course on lock-free data structures? Maybe also a course on succinct data structures 😉? Also nand2tetris doesn't cover GPUs AFAIK. But there are a lot of other examples.

Andrew DrozdovAndrew Drozdov@mrdrozdov

@srchvrs I mean. A decent amount 😅 I did take courses for all of those topics at some point. And something like nand2tetris does cover a lot of the CS stack. My hot take here is that AI infra structure in particular will see an explosion of complexity, although maybe I'm overestimating.

3:31 AM · May 27, 2026 · 73 Views
4:03 AM · May 27, 2026 · 42 Views

@srchvrs I mean. A decent amount 😅 I did take courses for all of those topics at some point. And something like nand2tetris does cover a lot of the CS stack. My hot take here is that AI infra structure in particular will see an explosion of complexity, although maybe I'm overestimating.

Leo BoytsovLeo Boytsov@srchvrs

@mrdrozdov We are far beyond that point. For example, how much do you understand compilers, cryptography, or complexity theory? Are you an expert in C++ template-based programming, do you know lock-free data structure?

3:25 AM · May 27, 2026 · 314 Views
3:31 AM · May 27, 2026 · 73 Views