There's gold in those old papers, and there are opportunities to use the old magic, today.
Perceptrons were concieved of in the 1960s!
Even after LLMs, developers STILL don't have a sense of the sheer scale of what a GPU is capable of parallelizing.
I can think of a half dozen innovations gamedev's came up with in the 90s that could be enterprise B2B SaaS products today.
Before I go on too big of a ranting tangent. Back to the original article: I'd still use the technique in the article in unity way before I'd use unity's built in terrain system. Wouldn't work great on mobile, but for every other platform it'd work great and could be implemented trivially AND I could write semi normal shaders for it.
I think raymarching signed distance fields for terrain is another interesting technique worth looking into. Costs a lot of fill rate, but it's got interesting properties.