The Ramanujan Machine
ramanujanmachine.com
ramanujanmachine.com
But note that it just generate special cases of a well known formula. More details in "The Ramanujan Machine is all hype" https://www.galoisrepresentations.com/2019/07/17/the-ramanuj... (HN discussion https://news.ycombinator.com/item?id=26144665 (40 points | 4 months ago | 16 comments))
If the problems being solved are useful, they are potentially more useful to some parties than others, meaning some parties may have a lower cost to attack the blockchain, because it was computation they needed to spend money on anyway, might as well attack a blockchain while you are at it.
I can try to explain further if that's confusing.
So for this to be a good computation for a blockchain to use it would have to have a way to modify difficulty in a predictable way.
What reasons besides predictable inflation are there for desiring that property?
Blocks being mined too slowly is a more obvious problem: less transaction throughput.