01:08:51 forgive my crackpot theory here, but 01:09:02 in randomx, all data is valid code, right 01:09:14 genetic algorithms work by taking data and combinining it 01:09:39 say you have a function that computes some difficult mathematical function 01:09:46 inputs go in input registers, outputs go in output registers 01:10:10 goal is to optimize it 01:10:23 could you do a genetic algorithm that breeds the randomx instruction sequences with each other? 02:25:27 sounds possible sure 10:36:43 A genetic algorithm would need to work on the input data to randomx (block header, variable part mostly the nonce), not the instructions resulting from that input 10:38:58 And generally the idea behind evolutionnary algorithms is that two closely-related sets of input data will produce closely-enough related fitness when evaluated 10:40:21 So for instance if you have a good parent A and a good parent B, you can combine them and have some reason to think that their offspring is also of interesting quality, and not just randomly bad like any random solution 10:41:41 That doesn't hold true at all for common hash functions, since one of the design objectives of such functions is that the output is "unpredictable" from the input, so there is no usable correlation between input and output. 10:42:48 You change one bit only to your input and the new output is not only completely different but also (hopefully) unpredictable from the previous output 10:44:26 That also holds true for randomx so I don't see how that would work unless some structure in the search space becomes exploitable 13:01:43 binaryFate: No, why? Couldn't you just work on the uncompiled instructions? 13:02:10 and keeping the input unchanged? 13:22:02 IBM showed off a 2nm process for CPU https://searchdatacenter.techtarget.com/news/252500291/IBMs-2-nanometer-CPU-roadmap-heats-up-chip-manufacturer-race 13:22:28 they're running out of nanometers already, only 2 left... 13:24:13 heh 13:26:51 Common misconception. You don't have to buy a whole nanometer, you can buy fractional parts of it, so if the price of a nanometer goes beyond what you can afford, you can buy 0.1 nanometer for example. 13:27:04 Wait, wrong pattern match ? 13:54:05 seems like 2nm will be the limit for silicon. after that will need a new chemistry, maybe graphene 14:00:26 yanmaani: You can play with instructions but you wouldn't know which nonce would produce them 14:06:21 hyc so perfect time to tune RandomX parameters is when 2nm chips come out from all major chip makers 14:06:31 sounds right to me 14:06:39 Moore's law is dead 14:37:28 moneromooo: my wife will not be happy with 1 nm 15:31:05 hyc: input becomes whatever the input to your function is. if you want to approximate "f(float, float, int) -> int, float", just start off by loading the inputs into say float registers 1 and 2, and integer regiser 1 15:31:30 and let the outputs end up in some other specified register 15:31:43 binaryFate: correct. goal is just to find instructions to generate a pure function