undefined | Better HN

0 pointsscotty794y ago0 comments

Sure. That's also great if you can monetize trained ML models so easily that everyone can participate.

It's already being done to some very small degree by few companies that locate their ML server farms in places where energy is cheap and renewable.

I hope this application will some day compete with crypto for computing power and energy and win but I don't think it will be popularized and commoditized soon enough to bring money into green energy.

I hopefully await headlines "ML training now uses as much energy as Australia".

If you could create crypto based on ML training it would be so cool.

0 comments

xiphias24y ago

,,If you could create crypto based on ML training it would be so cool.''

That would actually be a bad thing: the real value in proof of work is that it doesn't have any extra value to the miner, which stops miners from gaining a huge edge over other possible future miners. What's behind all the controversy is that proof of work has to be totally useless work, that's what makes it valuable for the blockchain :)

scotty79OP4y ago

Could you direct me to more elaborate discussion why work in proof of work must be useless?

Am I wrong thinking that any problem with hard to find, but easily verifiable solution, should be ok? It would be nice if it had easily adjustable difficulty to scale it with processing power of the miners.

1 more reply

intotheabyss4y ago

In proof of stake, you can slash validators that attempt to re-org the mem pool for MEV. Fast finality of a few seconds also helps.

j / k navigate · click thread line to collapse