@aburka yeah, that's true, the means to defeat anubis is not technically challenging
one potential explanation is that anubis is misconfigured, but another more interesting explanation is that an LLM company just did an utterly shit job at deploying their hellscraper (probably vibecoded it) and it's chewing up tons of expensive CPU and no one's noticed
or more sobering: they noticed but they don't care because their investors gave them infinity dollars to spend and it's not worth their time to teach the scrapers when to stop
@technomancy @aburka@hachyderm.io
Well, the idea behind proof of work was supposed to be that even if the bots start to do the work, that doesn't mean the system failed.
Anubis uses sha256 , i think because its the only one that's in web crypto, but sha256 is about the worst possible choice for this, its extremely easily parallelizable, tons of asics already exist, etc. So I would argue that sha256 is just a placeholder for whatever actual proof of work is used in the future.
That's why I started mine with scrypt, which is memory-hard, and I turned up the memory usage considerably. That way scrapers would actually feel the pain if they decided that they were gonna just solve it.
And I suppose with all these high-bandwidth-memory GPUs, maybe that pain wouldn't be so bad after a bespoke solution is developed. But in that case, I've already cost them way more time and effort than I had to put into defense.
And if I see bots solving the scrypt proof of work, I can just change it to Monero mining, lol. Maybe try to suss out access patterns and categorize human versus bot users that way, and then turn the thumb screws on the bots, make them really work for every byte. Or just send them into the iocaine Markov chain hell.
