Sort by date <https://hn.algolia.com/?query=low%20background%20steel&sort=...> and skim through and you’ll see it’s way lower. The considerable majority of those matches are about actual low-background steel for its own sake, not for analogy with pre-LLM data (or once, https://news.ycombinator.com/item?id=32097752, pre-microtransaction games!). Nuclear-related topics come up pretty regularly on HN.

Five pages of results; past the first page, there are only five relevant matches. https://news.ycombinator.com/item?id=23895706 three years ago (when it wasn’t yet a problem, but it was obvious it could become a problem), then the rest all within the last year (mostly this year) (when it is obviously becoming a problem), about thirty in total, half of which are multiple related comments. So the true number of times the prediction or analogy has been made is under twenty so far, and they’re mostly saying “like someone recently said” rather than coming up with the analogy by themselves.

The number seemed unrealistically high and I was curious. Now I have an answer! :-)