I've never seen an application without limits. The NIH set limits in the 90s and then further reduced them in the 00s.

I think ironically enough, the AI might actually be the tipping point that forces grant application to put tighter limits on length. Im a part of evaluating compute grant applications, and we are now getting extremely large applications regurgitating completely irrelevant information (some of which i recognized, because i helped author some best practice guides in the field). Its straining our process because we can't ethically offload evaluating the applications to LLMs. I suspect this will be a continuous theme of time; that LLMs will highly strain all human processes that for any reason cannot be offloaded to LLMs.