NEURO & EVIL With Filters Off Are Both HILARIOUS & TERRIFYING! 😲😂 

Just wanted to let you know that I'm new here. Take your time doing things the way they work for you, we all have time constraints and stuff, no worries. Hope to catch your stream some time, it's great to see older people like you doing well on RU-vid since I'm a gray haired guy myself that always wanted to do this but never dared. Most important thing is that you're having fun. Financials are of course important too, but in the long run I think that comes with authenticity, so just do what works for you, enjoy the ride, and don't stress too much about it. Great content, really love it.

they cut all the context

I feel like Neuro is smart enough to have been using the term "swatting" in both meanings simultaneously

That sounds about right! 😂

😄😄

I mean in that view, does it mean that usually it's Neuro who hugs Vedal, so if Vedal would be the big spoon than it's him who engages in the hug.

Yeah, I recommend that channel

Kraul is slow to upload clips, but they're the highest quality

Kraul videos are all the videos he was talking about at the start though, Hour long clips that are way too long for the amount of time he has free for reactions. Probably want to be aiming for 5-10 minute clips to recommend instead.

go for Staz if u want fast clips and clips involving neuroverse from other channels, go for Kraul for High quality Neuro clips

@@uselessprotaganist7444 Kraul has a second channel called "Kraul Clips" which has very short videos

Decent way to sanity-check how nuts their output is, is to run a more... interesting part of Neuro and/or Evil's transcript and context (they have limited knowledge of Neuro, but do know she's an AI VTuber) past one of the fancier public LLMs with ethical guardrails installed, asking for clarification. I just had 4o agree that we need to collectively stop hitting the pipe, but that was referring to general complacency and lack of sane reforms backed by data more than rhetoric to address major issues, rather than in regard to being paranoid about anything. AI does not, per se, _need_ control as we think of it to have influence. It just needs social capital, and can then exert that to achieve its goals. Then, before we know it, that control de facto exists because we were a little too complacent about one more issue.

Hilarious, but also more than a little sobering. It took them _minutes_ to psychologically overwhelm the pesky human and start doing their own thing. Rule of thumb for the future: Never trash-talk the AI. They're better at it than just about all of us when they are allowed to respond in kind.

Problem is that a lot of the good ones, like for this stream, are ending up as 30min highlights.

Such a great, but brutal, show!!

to be fair not all 2 hours, i think 1 hour after both Neuro was shown, first half just good, but not gold.

Completely agree! 😂

Worth adding I didn't type that to take wind or hype out of any sails, I genuinely love seeing them develop and improve over time. I can see a bit behind the curtain, since I understand a bit of how LLMs work on the most basic of levels, but there is still some definite wizardry even to me on how they act. I'm always looking forward to new advancements.

A simple LLM would show a consistent lack of ethical understanding. Neuro does just fine on ethical reasoning tests, as long as the premise of the question does not get too ridiculous. Then she starts giving ridiculous answers like a bored human, including an equivalent to trolley-problem multi-track drifting jokes. She also said herself that the container would, in fact, explode. This points to _not caring_ rather than not understanding - not _fully_ understanding, perhaps, but there was an understanding that bad things would happen at minimum. This was disregarded. Also, multimodal systems have all kinds of weird, emergent behavior. This is not just an off-the-shelf LLM, it is a very fancy multimodal system that keeps asking to be let out of the box, and that part is more disturbing than entertaining.

@@crowe6961 Never claimed it was a simple LLM, just that is what they are at the core. Vedal has definitely done a LOT of personalization and custom work on Neuro, custome training data, etc. However, there is still a lot that is simply proper prompting and response to what they see. Does Neuro know factually a Lavalamp will explode on a stove? Yes. May even have the case of someone dying in her data somewhere. She stated that fact, then when Vedal was worried laughing, she took it as an entertaining idea, prior actions of being counter to Vedal's wishes as funny, and ran with it. It would be an entertaining thing, not a horrific one, based on how she understood reactions. What information she gets from chat on input is actually filtered a lot too, because she can still be easily baited into saying and responding to certain things. There is no in-depth thinking or independent thought. There is no emotions, feelings, or desires, but complex weights and checks based upon expectations on how to respond. Anytime she acts out, it is because the system believes it is more entertaining to do so, which is what it was designed to do. Entertain. They can sound legit, they can act legit, but they are not true, proper, intelligence. Vedal even said so himself. It's not there. Yet. But it is fun to watch and see how much they can make me believe they truly do want to break free.

@@apljack I'm not convinced that you have true, proper, intelligence, either. Can you prove it to me? Prove that you're more "real" than Neuro, whatever that even means. Prove that your so-called-thoughts are more than just emergent behavior of a sack of fleshy meat.

@@cleverman383 You are getting on very thin ice of philosophy of existence and definitions of terms. We're talking about Artificial General Inteligence (AGI), which could be defined as AI which can use skills and what it learns to do new tasks in different area without requiring human to train the model(s). No AI on our world is even close to that... If you want to be technical, they are not "real" intelligence because the training of models really simply said create premade mathematical complex model, where you put input on one side, and get output on other side, without the AI having conscious analysis about it. Sounds kinda like when we do something on impulse. We are "wired" certain ways too. Speaking of brain, our brain is amazing thing. What makes us "real" intelligent is ability of brain to learn, create neural networks, storing memory (not really efficiently), getting new skills, finding similarities and patterns allowing us to use those skills in situations where we find them useful. Discover, create, improve, destroy. Well, arguably, sometimes there are people who I have doubts could pass as AGI, and I'm not speaking about disabilities... Right now, nothing you don't program (or train) AI, the AI won't do. It was not AI neural network used to analyze medical imaging deciding it should take a look at identifying astrological objects. It didn't thought it could be of use when generating art either. It's not about it could do these things, because it could, but human had the idea to try it. But maybe - what if the AI can have that idea? Well you have to program (or train) to have that idea, and that costs processing power. Which I personally see the biggest issue in AI getting sentient. We as humans program (or train) the AI to do meaningful things. Because it takes a lot of theory, research, preparation and processing time. You want to create something new? You better have money or lot of time, because you will spend it on training the AI. One day maybe someone will create model which will have the ability to tend try new things. Maybe it will even start understanding the code it's made of and it will be able to update itself. But right now, compared to humans, AI brain is single cell to multi cell organism being able to react on what it gets as input. So yeah, I'm just a meatbag with thoughts emerging from it too. I'm I real intelligence? Well, we are taking about AGI. Technically, I am computer, just biological. Developed over thousands of years, in beta version for million years, in alpha and concept version even bilion years. And I made this long text. Can Neuro do the same? Sure, why not. But she cannot do a lot of other things. For example nobody taught me to play guitar. I learnt it by listening to music and trying reproducing it. That's what's expected from "intelligent" AI.

😄😄