Doesn't mean they didn't apply something similar. They could have also come up independently with their own version, the speculation is not they copied it, rather that they have performance breakthroughs which perhaps is a result of work in same domain

Models are generally posttrained to a window shorter than you think.