Knowing the answer isn’t really useful - for reasons you’ve pointed out - and setting records like this is always just done for the publicity (if you wanted more digits, you can just keep the system running the calculations for longer).

Doing the calculations themselves does have some uses though - it turns out that its actually pretty hard to find problems that can totally saturate the CPUs of high performance systems for long enough that you can see rare issues that only occur under significant load.

For high end compute stuff, you need a task that you can run in parallel across hundreds or thousands of machines, that isn’t going to be bottlenecked by I/O easily, and where the results are easily verifiable - so you know the performance and reliability characteristics of your system before you commision it to run real work (where you operating costs might be $10k/hour+, and there is a large queue of work waiting to go)