On one hand... the idea that tables used for representing tabular data and tables used for layout couldn't be distinguished and managed functionally is odd. Lynx could handle table layouts reasonably well by the end of the last century, and I don't think you'd even need an ML model to code up some reliable heuristics and get them into most browsers, though I expect ML could be used to pretty good effect.
On the other hand, getting universal buy-in on a different approach to semantics and ubiquitous client adoption is harder than just buying into the agreed on semantics. And that recognizes the opposite of the hacker spirit. "Real hackers", if that term means anything, recognize the value of adhering to semantics by convention in all sorts contexts, even lisp-y ones. And while I have no idea what other things Arc's web library does, it's pretty clear to me that there's little if anything in HN's layout that wouldn't benefit from using CSS over tables for managing layout.
On the other hand, getting universal buy-in on a different approach to semantics and ubiquitous client adoption is harder than just buying into the agreed on semantics. And that recognizes the opposite of the hacker spirit. "Real hackers", if that term means anything, recognize the value of adhering to semantics by convention in all sorts contexts, even lisp-y ones. And while I have no idea what other things Arc's web library does, it's pretty clear to me that there's little if anything in HN's layout that wouldn't benefit from using CSS over tables for managing layout.