LLMs are very good at taking your existing structure and converting it as-is to a new language. But rewriting by hand gives you the opportunity to rethink previous design decisions or optimize the code for the new language. So I was disappointed that this appears to just be a straight port and no additional performance improvements were gained.
"Recent advances in AI-assisted coding have made large-scale rewrites considerably more practical, but they do not eliminate those risks, and we do not take them lightly."
Last time I checked it didn't make any difference or whatsoever when compared to lld so I would not go that far by saying "mold is a reference in the world of linkers" - made a test just few days ago on 10M LoC C++. So, rewriting the codebase in another language just because doesn't seem like an investment worth doing when there's much bigger fish to fry.