>>54675 Biological understanding is different from logical understanding, and no one is trying to make AI that works the same as humans. The key is emulation. >we don't even understand how humans think We also don't understand how current AIs "think".

i wish i knew less about AI than i do, so i could dismiss the idea super-intelligent machines in my lifetime with the same simple minded certainty

>>54704 do you actually know the statistics parts or do you just fuck around with the generator stuff?

I watched episode 1 and it felt like I'm missing out of an entire season worth of story and left me greatly confused. Is it just some kind of "in medis res" story telling that will eventually catch me up on everything? I swear I don't see any mention a previous season.

>>54725 well the brain diving shit is sort of thrust uselessly into the plot a bit but the details of human brain implants are trickling in through the episodes and are probably thematically in line with the android AI bits.

>>54725 Anons back at 4chan said that they skipped the entire first volume I think

>>54742 But like who was that girl she was shopping with at the very beginning? How am I supposed to know what a royal model or whatever is? Why are they in Russia? Just lots of things dumped with no explanation as if I should already know. >>54747 Okay that would make sense as to why it seems to be in the middle of things, but it also raises a question of fucking why skip a volume?

>>54753 the girl at the beginning is harold's owner? host family? or is somehow associated with their family. RFs seem to be exclusively british so that's not explained yet, but the st. petersburg location is some major plot incident that has been hinted at heavily. a royal model is a model for royals, they recently got up to talking about The Queen in the latest episode. it seems very unimportant except to bring them to london to jerk off over england and oxford and sheeps.

>>54628 Dude, we don't even have a rigorous definition of AGI yet and you're saying it's close to being achieved.

>>54704 it's specifically what I know about AI that lets me write off AGI with a "never ever and anyone who says otherwise is a grifter or an idiot" but if you know more than me then go ahead

>>55212 >it's specifically what I know about AI that lets me write off AGI with a "never ever and anyone who says otherwise is a grifter or an idiot" Explain further

>>54709 >>55212 Given that the bar for me to clear is pretty low - prove that the idea merely doesn't deserve to be dismissed out of hand as opposed to being bullish on specific timeline, it's enough to say that: 1.Few years ago we had a paradigm shift no one saw coming and radical expansion of what AI can do. 2.We are still in the middle of it, not even all of the low-hanging fruit is plucked, throw more dollars and compute and get better results, easy as Moore's law still works, btw 3.we don't really understand what intelligence is about, hence being blind sighted by the fact that you can just put together some basic ass algorithm, feed it unholy amounts of data and get software that exceeds at stupidly wide range of tasks. Seems like grounds enough to refrain from confident statements about exactly what happens next. Taking more birds eye view, there's no reason to even think there's some kind of profound barrier here. Biological intelligence is constrained by neurons being orders of magnitude slower than silicon, nature working blindly with no foresight and 'aiming' at survival rather than wits, the width of your mother's vagina, the list goes on.

>>55401 bro its just text inference

>>55412 Yes. Hold onto that thought when you're being escorted to biodiesel conversion factory by your new robotic overlords for additional sense of humiliation

>>55308 models are a closed system that infer from parameters set during their training phase and are incapable of producing any output that falls outside of their dataset they are trained on a set of data that is flagged by humans as "good" or "bad", but they cannot conceptualize what "good" or "bad" mean except by association with the previously set data if you try to write an algorithm that will train its own model based on rules as to what kind of data is good or bad, then the problem recurs because they have no capacity to understand what those rules mean and are bound to make false predictions based on edge cases it hasn't seen before this problem compounds in a phenomenon known as model collapse where the model is polluted by data that it can't determine is good or bad and it ends up outputting normalized garbage therefore AI models are incapable of truly learning in order for that to change and to reach AGI their fundamental status as a program written by humans would have to change >>55401 this is literally just hopeful ignorance and nothing substantive is being said, you'd probably reply to that last line with "they're working on it, it'll be ready in two more weeks!"

>>54974 We've had a general idea of it since the term was invented at the turn of the century. In fact the meaning of the term has been constantly updated and debated recently since the strongest modern AIs can already fall into the original definition. It's increasingly becoming a case of ///DOESN'T COUNT\\\. If you've been living under a rock recently you can see this happening from the wiki timeline alone: https://en.wikipedia.org/wiki/Artificial_general_intelligence#Timescales >>55588 It's amazing that you can write all that yet miss how that is exactly how humans learn from culture during our childhood. >an algorithm that will train its own model Pretty sure this is already what GPT models do. The developers just add stuff to prevent it from generating tutorials on how to make bombs or something otherwise illegal.

>>55588 >are bound to make false predictions based on edge cases it hasn't seen before as opposed to your average human? None of what you brought up is some kind of fundamental architectural limitation. Feed your reply to chatGPT for detailed breakdown of current state of research and how these problems could be addressed. Again, current architecture is fairly simple because a range of abilities basically just emerges on its own with enough good training data and compute. I imagine you're so smug because you saw this coming the entire time? > hopeful ignorance and nothing substantive is being said while your chatGPT tab is still open, can you also ask it to tell you how ironic this statement is in light of the rest of your reply? >you'd probably reply to that last line with "they're working on it, it'll be ready in two more weeks!" I'm definitely not "hopeful" for machines rendering human wits obsolete. and I will say again that I neither argue for nor believe in "AGI just around the corner" You just insisted on picking an awkward hill to die on. If I was talking to an LLM, it would understand this much easily btw, so this has some grim implications about you.

>>56925 an entire post of nuh uh you actually don't understand anything about the underlying technology if you think any of this is subject to change >Feed your reply to chatGPT for detailed breakdown of current state of research and how these problems could be addressed you realize that by doing this the best possible outcome is that it regurgitates an article that entered its training data written by a human about this issue, and it couldn't possibly act on what is proposed since it doesn't know what it proposed since it doesn't know what words are

>>56944 >it couldn't possibly act of course it can't, it has no body, not much context length, no memory, can't properly modify itself and in the first place it just wasn't build with that in mind. will you insist that any of this is insurmountable? even llm can think more steps ahead than you. > it doesn't know what words are A2:it turns out that you don't need to know what words are to write essays, poems, code and billion other things. and neither do you need to know what objects are or how anatomy works to draw just about anything. maybe it's time to reconsider what words like "know" and "understand" actually mean

>>57019 just fuck off you clueless retard this is why sci-fi threads are always dead btw

>>57051 lol your concession is, of course, accepted

>GPT-chan give me some realistic looking Python AI code >build >jew runes not in a string Also cyberinspector continues to get manhandled. She really should undergo some kinda training so that a computer nerd can't just slap her down. Also look at how goddamn flat she is fuck that is hot. >>50996 They're like the fucks who came up with Roko's Basilik, having so little to worry about they need to come up with abstract worries. Hey, imagine if in the far future a hyperintelligent AI actually came about and that it would be so petty as to look over everything it knows to punish anybody who might have slowed its arrival. Right. That won't happen with ML, I can guarantee you that. Computer nerd's worries were that given they weren't mimicking thought, they could develop and become amoral. His sense of scale is wrong however, humans kill more humans everyday than those three robots ever could. >>55401 ML is not AI. ML is pattern matching and linear algebra juiced up 10000x. Neumann had this nice quote about fitting curves, which in the end is what ML does: "With four parameters I can fit an elephant, and with five I can make him wiggle his trunk." Besides, if you outside the test space, ML models' accuracy drop off like mad and it literally can't self-recover, unlike you learning not to touch a hot plate. Literally unsuitable for something for intelligence.

Is all interest lost in this anime? Today's episode was chaotic and depressing. The OP and EP were the best moments...

Sex with Echika! Don't let a thieving cat steal your android from you! E-drugs are still made by Walter. Also fuck these always-hooked-to-the-net cops. Suspicious solely for not wanting to be surveilled 24/7? E was right all along! >>63715 Dead board is dead.

She is so flat holy shit Yuck

>>63720 It was funny that Echika could ask just any random person for documents without a reason. But turns out he wanted to kill someone, so everything is justified and no privacy is required. Meanwhile, Bigga the bio-hakka is distributing untested chips for your cyberbrain that may help you or may waste you forever. I still wonder if that affected Echika. >>63726 Echika body is perfect. Flat and legal.

Got caught up. Watched the 5 episodes back to back today. And man, was it boooring. Legit, I was yawning at it. So many exposition dumps.

Is she really a world class investigator? 6 episodes in so far and the only pro thing she has done is brain jack into people for like a few minutes. Other than that she doesn't really do a good job during any of the investigation scenes.

>>65894 they tracked someone through unfamiliar foreign wilderness and captured a supervillain

>>63715 sexo

>>65894 It's more about her innate data processing abilities. The cyber investigators don't really do any real-world investigations, they just brain dive.

>>65917 there is some sort of memory-navigation aspect to it that is like meta-psychological or something, not well defined in anime.

She's beautiful

Those fucking luddites went TOO FAR now. >>66043 But also hungry for robocock.

What kind of retarded fire escape doors do they have? If it can already be blocked off by something doesn't look like it'd weigh more than 10kg, then there's something seriously wrong with it. Plus, it's even inside the building so you could argue for doors opening in both directions. Also LIAR LIAR BACK ON FIRE Echika and Android should just get a room already.