"all chat g-pee-t can do is use water to steal shit and lie"
AI researcher from SF on the precipice of the kurzweilian singularity witnessing the transcendental hyperobject at the end of time self-assemble itself in O(logn) time:
DeepMind's Andrew Trask on why cryptography keeps beating AI:
"Cryptography and machine learning are kind of opposites. They're a yin and yang. Machine learning is like, no matter how noisy the signal is, I'm going to recover the structure and figure it out. And cryptography is like, no matter how big the structure is, I'm gonna hide it in this noisy thing that provably you can't find unless you have the key."
"So they're arch nemeses of each other, locked in an epic battle to figure out who can hide structure and who can recover structure."
"Up until this point, cryptography has won overwhelmingly. There's been a long-term defense advantage that's come from cryptography, and this has had enormous political implications, the ability for everyday individuals to send messages that no powerful state or centralized power can see."
"The real problem we're facing is not a weakness in cryptography. It's human flaws in the proof-making for new things, and the rollout of distributed systems, getting people to update versions of software. That's a very hard problem, because it's a collective action problem, but it's different than cryptography being broken."
"In the end, cryptography will win, but that doesn't mean that no one will get hacked. The cryptography is not going to lose, but the people holding the keys to cryptography are vulnerable. So what we actually have is a people-versus-people problem."
@iamtrask@openminedorg
@Esoteric_Though@Polymarket Shor's Algorithm is PQC? You must mean that current algorithms are quantum, for current algorithms to be post quantum we have to all be on post quantum algorithms. Anthropic is fearmongering a trivial research find. please stop being ignorant and sucking their balls
The only flaw is that an inevitable advance in math from a combination of AI and quantum computing forcing the algorithms to grow in a linear scale to match and counteract the future compute power that will be available. to say that Cryptography right now or even less PQC is vulnerable right now is saying that we have cracked pandoras box and we are going to become a tier 1 civilization soon after.
open source ai is a must have, to combat bad actors with frontier level agents, good actors must have access to the same caliber. it is the equivalent to gun safety laws. the criminal will find a way to acquire that gun (jailbroken frontier model or open source chinese models) the innocent need ways to defend themselves (frontier models being open sourced)
@Polymarket open source is needed to combat bad actor agents. guardrails are like gun safety laws they let the bad guys do whatever they want but the good guys have to find alternate means to defend themselves with.
also it seemed snarky my reply, but in reality the main worry right now is everything that can be harvested now and then decrypted, so it is more of securing everything as much as possible first, because bad actors who acquire your encrypted data will eventually be able to decrypt it, yes recommended to move but HAWK has just been cut in half by PRE quantum computing AI. yes recommended to migrate of course, pqc is the new best thing but it isn't as established yet. standards are set but with AI ever evolving as we can see here, it is inevitable that it will take longer to find a proper standard as institutes finalize everything they want it doesn't mean their word is law. and the timeline is around 2030 as well, so still. not worth fearing it yet or fear mongering. just another year 2000 fright.