anthropy

anthropy, 9 days ago

@VonKarth I built my own infra from scratch, Currently mostly using Fedora and AlmaLinux, but my stuff is distro-agnostic and works across linuxes.

I have 2 servers at home (~80tb of storage, ~512gb RAM, lots of both AMD and Intel cores), but I only have 200mbit up, so most of the frontend and data-heavy stuff is on some dedicated servers at Hetzner, though again those just run my own software on that.

I mostly use Docker and Podman, but I'm working towards agnostic container infra in general.

anthropy, 9 days ago

@VonKarth hmm, bit weird of a subject still because you're still trying to indirectly complain about modern society rather than actually asking an open minded question about AI and quantum systems.

But whatever here goes; Society currently seems to get upset about "AI" in general with arguments like power usage, when it might as well be used to REDUCE power usage and improve life for everyone objectively. The fact people try to see it as some kind of infallible entity is causing disconnect tho.

anthropy, 9 days ago

@VonKarth Quantum computing has decoherence issues, that is to say, a qubit doesn't accurately retain its values for long, so the only computing you can do has to happen within milliseconds or at very least seconds of storing the data in the qubits. There are plenty of examples that work with this though.

anthropy, 9 days ago

@VonKarth every AI architecture, training, model, etc is different, and the biases are also different as such, every entity that exists is biased, also in its ability to understand the world based on what it has been exposed to; that is inherent to existence and hardly anything that can be attributed to AI specifically.

if anything, most LLMs currently do not actively reason, so they may not pick up on sarcasm, but they can, and the rate of accuracy is not super far off compared to humans tbh

anthropy, 9 days ago

@VonKarth LLMs aren't really mimicking data, it's not like cleverbot or alike that just copy paste inputs, it creates billions of parameters to deduce what kind of responses make sense in a certain context, it doesn't even remember full words, only parts of it, it's actually more like human reasoning in some ways.

You don't need quantum systems to produce AI but you can totally write transformer style AI for quantum computers, it won't make it any smarter though, that just requires arch changes

anthropy, 9 days ago

@VonKarth 1) wikipedia is human written and usually refers to sources to verify your answers 2) it is definitely true emergent intelligence, it's just not perfection because again, anything that exists is only perfect within a certain context. It'll simply improve more and more as time goes on though, I am entirely convinced LLMs could evolve into true independent agents one day, as much as humans like to think they're special somehow, they also love to neglect the intelligence of animals, so eh

anthropy, 9 days ago

@VonKarth we have easily enough normal computers (like the TPU pods) to simulate multiple human brains at once, but just simulating neurons doesn't make things smart, quantum computers make no difference there, they're just a different way to do the same math.

LLMs already use deep reinforcement learning, chain-of-thought reasoning, and so forth. LLMs are smart, I think people are just afraid of looking in the mirror and realizing they are already much more stupid than the average LLM :P

anthropy, 9 days ago

@VonKarth I'm pretty sure if you go across a wide field of sciences you'll find that the amount of information you can get out of an LLM is far bigger than even a smart human. Is it perfect? No but you can bet that the human also has areas they will make mistakes, not to mention they have to look up things, which LLMs can also do, but don't need to, because their memory is much better than a humans, especially if we're talking about the ones with massive 2+ million context tokens for reasoning.

anthropy, 9 days ago

@VonKarth

quantum computers are not special magical entities, they exist already and you can already build AI with them today, the results are really not as exciting as you'd think

https://quantumai.google/cirq here go play with 'quantum ai' :P

anthropy, 9 days ago

@VonKarth not exactly, you do not have 2 million context token short term memory, brains have a lot of small interacting systems that give rise to a 'sense of self' and such, but that's mostly just due to our sensory experiences, but I entirely expect even the most basic LLMs to completely surpass even the smartest human within the next decade.

We need to stop comparing and just work with who we are, there's absolutely zero sense in trying to compete with technology, like trying to outrun a car

anthropy, 9 days ago

@VonKarth they're not faster either, they're good at factoring specific things and are efficient in that sense per qubit compared to per bit, but you require far more power per bit because you have to cool it to almost absolute zero for it to work at all.

anthropy, 9 days ago

@VonKarth like really, you have to go look into quantum mechanics, and why it's called QUANTUM mechanics to begin with (quantum being a measure of something, an amount), it allows you to express potentiality waves and do weird things, but they're actually kinda bad at classical computing and should not be compared to that at all

anthropy, 9 days ago

@VonKarth I mean that's fine in itself if you don't say things like "but they ARE xyz" or "they do abc", just add "I think" that would go a long way, I think :P

quantum mechanics, chromodynamics, field theory, etc are all fascinating fields, I can much recommend looking more into it, and actually trying to understand it rather than how spooky mysterious it is where a lot of people seem to get caught up on.

anthropy, 9 days ago

@VonKarth no worries, glad to have some conversation that isn't about violence and drama heh

would also recommend talking with LLMs, they're not as stupid as you may think they are, and they are getting smarter every day too, having a lot of interesting things to show about the subjects, and can link you to the sources to read up on human-curated things

anthropy, 9 days ago

@VonKarth I mean I could link you to things like https://en.wikipedia.org/wiki/Quantum_field_theory but I think something like GPT4o or Gemini advanced would do a much better job at explaining it in ways you can understand

compare these two:

image/png

anthropy, 9 days ago

@VonKarth find better people to talk to? And holy shit stop expecting other people to fix you?! Stop being so obsessed with negativity, stop finding and starting drama, just have a normal conversation FFS

Your next @ to me better be about fucking servers because I have no patience with this left, I will mute your account next if it's literally any other subject.

anthropy, 9 days ago

@VonKarth x3 look dramallama, you're the one starting this, I keep repeating there are subjects I'm fine with, but you choose to get upset instead, and at this point I'm strongly suspecting you do that on purpose too.

Muting this thread.

anthropy, 10 days ago

@VonKarth lol, I guess science uses latin? politically there's not exactly much left of the roman empire lmao, it violently fell apart and then everything re-formed quite a lot of times before we arrive at modern day Europe, the only things left are artifacts like the Colosseum in Rome.

anthropy, 10 days ago

@VonKarth I think you're better off asking chatgpt tbh, I don't know that many ancient romans to set up an accurate objective comparison by any means :P

anthropy, 10 days ago

@VonKarth ChatGPT seems to understand it just fine, although I'm not sure how 'profound' I would call it aside from people that actually care about history beyond how to avoid the mistakes from back then in today's era. I suppose people in certain regions like catalonia/belgium(flemish)/etc might feel strongly about it, but again it's mainly the result of thousands of years of wars and capturing back and forth, you might as well be debating the influence of the spanish, or the vikings, etc.

image/png

anthropy, 10 days ago

@VonKarth I just kinda expect you to constantly try to turn it into something violent, given a lot of things you said in the area of politics before. I'd much rather talk something like computers with you so that you just cannot turn it into something like that >_>

anthropy, 10 days ago

@VonKarth I mean don't take this wrong but you've kinda not been giving me much reason to trust on those subjects before, so forgive me for being hesitant with those kind of subjects. I'd rather just stay away from those, unless we get some of that trust back, as much as the history of europe in itself in ancient times is a fine subject, I just don't want anything that relates to modern (socio)politics right now if that's okay 🤷

I'm glad to hear you're looking for ways to be less unstable tho 💙

anthropy, 10 days ago

@VonKarth It's fine to be more positive, but I'd rather stick to non-risky subjects for now, if that's okay. I'll try to let you know more clearly when I'd rather have a different subject instead of be what you perceived as condescending about it though.

anthropy, 10 days ago

also note that I named the KESSv2 and KTAG here but there are many similar tools, basically you want to talk with the ECU directly so you can set different mappings and extract the existing mappings, which is done over.. some protocol I'm still trying to figure out, seems to be CANbus/K-line/J1850/etc, but also some ECU/OEM-specific pinouts and things, some cars require an extra piggyback chip and won't let you direct access, but Renault seems to be somewhat lenient with ECU access in general

anthropy, 10 days ago

@puppygirlhornypost I appreciate how they carefully guided all the cables to the side but once they realized it was going to be way too much they just went back to The Pile ™️