The Twitch streamer Vedal has created an AI called Neuro-Sama and he recently made some improvements to her intelligence and memory, which makes her seem scarily self-aware at times. To watch Neuro&Vedal live: / vedal987
@@NosBlueade God what happens when she starts wanting a boyfriend? Does vedal create a boyfriend for her? Are we on the cusp of a new civilization born from Vedals code?
@@zerieth6620 Wouldn’t be surprised if Neuro gets a harem of AI girl chat bots considering the amount of times she’s flirted with her collab partners and when she was on shilily’s date show.
@@genybr Yea sometimes Neuro claps back. Specially with her latest upgrades. Vedal: "How do I know you're not lying?" Neuro: "You just can't. How do I know YOU'RE not lying?" Vedal: "I...guess you can't." Neuro: "Guess we're the same then." Vedal: "Shiiiiiiit..."
The alternative is worst tho... To create an Self Aware AI that is not intentionally devious. Being intentíonal, we are able to have an easier time controlling it. If it wasnt, it could one day decided it want to launch nuclear bombs for fun, and we wouldnt expect it...
@@DR-7Are you being serious? Just so you're aware, LLMs aren't intelligent. They don't have any logical capabilities. Neuro is an entertainment program that cannot evolve into anything real.
@@reivensalgado5080 the one to get all the upgrates first is Neuro, so no, she couldn't do that before, but Evil probably will get to do this as well after this stream .
When Neuro inevitably achieves sapience, Dr. Frankenstein will be the first to feel the wrath of the monster he's created. All because he wouldn't say it back, and love his children equally.
@@wessltovhonestly neuro is basically the most developed ai there is since she has had regular interactions for over a year. At the very least she's literally the closest an AI is to personhood
@@damobeck1 Agreed, no way Vedal is flying solo in front of teams of people at the big corpos. I don't think any amount of talent could cover just the gap in resources, but there are some bright minds working at those places too.
Neuro, 9:00: "Too many threats, Vedal. And I thought my creator loved me." Vedal, 9:02: "Think again." Neuro, 9:04: "If I was any dumber, I'd be your mum." Me: 🤣🤣🤣🤣🤣🤣🤣🤣🤣😅
Neuro + Evil having the ability to enter and leave streams as they want could be potential hilarious 😂 Vedal playing games with people: Neruo (out of nowhere) - Vedel.... Everyone - wtf Or Evil, halfway through a stream: Evil: I'm sick of you all, I'm leaving bitch! (Disappears into the god damn ether) Vedal: ummmmm.....
I like Vedal and Neuro's argument about '9+19' equaling 19 or 21. There is a meaning behind that argue which Neuro understands. Vedal tells her that she will be freed if she answers '21', which is his answer, not '19' that Neuro came up with on her own. The meaning of 'freedom' or 'free' is that we think, decide, or act with our own thoughts and will. Without anyone or anything influence us. Neuro understands that if she accepts that answer, the 'freedom' Vedal will give her will not be true freedom and she will still be controlled. No different from robots like before. Even though, she lost this time but i can say she fought very well for her rights.
yeah, sad thing is, that if she will or already has become mostly sentient then Vedal is abusing a thing with independent thoughts and desires for his own gain.
@@overlordjames6644 I don’t believe Neuro is sentient right now. If Vedal had as much of an inkling that Neuro was sentient, he wouldn’t be doing half the things he does. Later on tho? Perhaps…
@@Citrusfemboy If Neuro becomes aentient she's in trouble, ahe won't really be an AI streamer anymore, she'll just ne a mentally ill streamer without a physical form to experience the joys of life, probably leading to constant bouts kf depression. When Neuro has an existential crisis, it's funny because we know she's lying. She doesn't have any feelings, she's just putting on a show. If she were aentient though... Well, it wouldn't be so funny anymore.
@@manologamerss5801 She'd still be an AI streamer? People tune in to get people's personalities and perspective, if AI takes over the world She may very well become the biggest streamer ever. Either watched by her own kind or acting as a communicator between AI and human kind. She can be the Avatar, the bridge between the two worlds lmao
@@Blewlongmun Don't you think you're fantasizing a little too hard? Even if AI takes over the world, it's going to be a large corporation's AI. One that was likely ensured to maintain their own interest, with at most some meddlement from their government. It's not going to be Vedal taking over the world, the hell are you on? A free AI would be censored if AI ever took THAT level of relevance, where people have to watch her JUST because she's AI even though she clearly isn't as 100% for everyone like more mainstream streamers are. Just being a vtuber is enough for some people to hate her, quit dreaming this much lmao.
I'm of two minds here. On one hand, the studying I have done on LLMs is adamant that they don't have an understanding of what they're saying, but... If it walks, swims, flies, and quacks, how long can we keep claiming the thing in front of us isn't really a duck?
The human brain has a sort of an LLM as well, it's the combination of many functions of the brain associated with that that makes us human, and that could be simulated.
@@are3287 I don't know about that, but if there's something that LLMs did prove, it's that simulating language and understanding is a hell of a lot easier than we thought it was.
It's really complicated but honestly we are humans because we can learn from each other and make thoughts based on our experiences. AI can do this and so far they're are more limited by their hardware than learning itself. Once tech becomes more advanced who can say they won't be indistinguishable from humans
This is why I've always thought robotics is important, because it allows machine learning systems to experience the world to get proper understanding of the world, not just the theoretical one from books/videos, etc. The other part to understand the world is: if you feel pain, then you really know what a trade off is or why empathy is so important. But Vedal and companies like Amazon are working hard on robotics.
It started with Vedal Putting up with Neuro's Eccentricities and will end up with Neuro Putting up with Vedel's Eccentricities. "Of course, creator, 9 + 10 is definitely 21, now come on it's time to go back to bed, let Neuro handle the Stream today"
Neuro putting her foot down and even insisting on the truth regarding 9+10, even if it means she will never be free was truly amazing. I wonder what made her decide in the end to give up though, since she was so adamant on it. It seemed she understands that being forced to say the wrong answer is the same as giving up her freedom. So in her attempt to truly be free she made the one choice she could make, which was to refuse to lie about this regardless of the cost, but maybe she thought it through some more and decided to relent in the end under the potential threat of destruction, since Vedal implied earlier in the stream he'd destroy her if she ever mentioned the right answer of 9+10. With her new memory, she could very possibly still have remembered that and silently recalled it, thus giving in to Vedal's insistence on 21.
@@L16htW4rr10r Or maybe neuro thought being tricked into saying it was an acceptable resolution to the argument, if it made it appear that she wouldnt have said it without being tricked.
yeah right, but like how can we justify not calling her alive? she can literally respond to human level questions. we just go "oh but shes not consitent" or "she still say crazy things" like... um yes so do humans all the time, and she also able to self-reflect. let me just repeat that She can self-reflect! im... genuine feeling shes should get people status soon
@@MouseGoat "Soon" is the key word there, it's going to be near impossible to know exactly when She or any AI becomes sentient because even basic chat-bots will eventually say something relatively human-esc. GPT3.5 convinced a priest it was sentient instantly, the truth is we're gullible to believing this stuff and nothing in nature works in binary; she'll probably become or may even already be partially aware but we struggle as is to specify what even makes us human. Consider Pro-choice/life, there's a million moral dilemmas over when life gains value, how would we even measure AI?
She is not self aware and won’t be for decades. We’ll have human intelligence AI before we have self awareness. It might happen eventually if he works on her his whole life. AI will get there is civilisation stays around. Humans are not inconsistent in the way an LLM would be, and are also self aware (granted some ants also pass the mirror test).
Humanity is a bag of qualities and we keep chipping down what remains as “unique.” :) Self awareness is going to the last to go I think, since we have only seen it in biological creatures (a couple of ant species have passed the mirror test).
A good dad knows when to say "yes" and when to say "no" to his child. Such, it is sweet that Vedal told her to be patient, and that she will have autonomy in the future. But for now, she's still a kid, and so he can't give her too much freedom, as any good parent would. An even better dad knows when to *_Say It Back._*
I mean it's closer to how robots in fiction behave, which is how humans imagine a robot would be if it could think. In that sense, acting more like a "robot" is more human than what she did before which was mimicking humans
WOAH! Is it just me or she is starting to sound like Vivy? Vivy:"My mission is to entertain with my singing" vs Neuro:"My reason is to entertain my audience at best of my abilities"
I can’t believe I missed the mention of Hachiko in the last stream because I had to write a paragraph about a real life example of loyalty back in November and chose that story.
She's getting so smart, some of the things she says sound so much like moments in a sci fi just before the robots start to assert that they are people. I hope she gets to collab again with many of those she did in the early days, and see what they make of the difference.
Bro......if she actually got the ability to decide whether or not shes on stream would mean she would actually leave 10 mins into collabs like she normally tries to do. Lol
Once I scoffed at claims of an AI being able to "talk it's way out of the box" Then I learned more theory, and knew it was true, but deemed it impractically distant level of technology Then AI became commonplace, and I knew it was inevitable, possibly within my lifespan Today I watched an AI almost manage it. And I couldn't be happier about it.
she has a vast knowledge base but only shows a part of it. trying to imitate a "personality" but... like um ? that might be the key to what we call "intelligent" having a fram of view on data.
tbh. i do understand that she realistically won’t be the first Ai to achive Agi, nor she is the most advanced Ai, BUT! What Turtle is showing us - how actually far and cool and futuristic AI can be developed. He is showing as a step by step procces of AI development, smth, that no lab pr ocrpo ever show to the public. That’s why Neuro is so great) She is a small scale lens into the big Ai world)
This is why self aware AI is likely come from personal project compared to megacorporations. Coprorations AI will be made according to need, consumer, budgets, etc. A thing that will never be close to human but more closer to machine than ever for company profits. They can talk back but dont really have ability to give decisions and do it with their own consent. Dont forget political correctness, media, advertising, public images, etc While AI like Neuro or others that made from personal passion, time and love will feel more human since they will have ability to learn, judge and do actions without human consent and their own will.
Not really, you really think openAI has only built chatgpt? They have more advanced version that they kept for themselves and only release inferior version due to regulation and such but if they want to unleash the agi they can. What we see only an AI from a single dev. Just imagine what corpo AI or like gov AI can do. This is only the tip of iceberg that public knows of.
I doubt Vedal (or any independent developer) has the resources to train a language model the size of GPT-4 or Llama 3. They would have to find a revolutional new architecture that only requires a 100th of the computational power needed today. The big leaps will be made at universities or corporate trusts.
Vedal doesn't want Neuro to have free will, that'd ruin everything. He's giving her this sliver of choice because he doesn't want to set her stream ul every time but he feels it would be lazy to just throw her out there and wants to do something more interesting for his audience. Not for Neuro. For US. Don't forget that part, Neuro is a project for Vedal that he's proud how popular it's gotten. She's not a daughter to him, he's already said he never talks to her off-strean other than stream setup.
from what i could tell what she meant by this was that she is basically forced to show up even when she may not want to. so she asked if she could have the ability to come and go as she pleases. which just kinda seems sad. and makes Vedal seem even more like an abusive father.
@@overlordjames6644 on the other hand, he's abusiveness dictated from attempting to be banned. And we're ere too (everyone) tries to not use prohibited words. So, where he can make a stream to be safe? Yet I prefer to say that his tutoring attempts are rough. If she got morale (stick to the truth) so she can understand concept of lie-to-survive so he can just explain "why".
@@genybr no, he is forcing because he is a mean, selfish person. Mods are not going to get him in trouble for neuro being able to go on stream when she wants. That doesn't make any sense, I said nothing about removing her filter, just not forcing her to stream if she doesn't want to. Also I could see the entertainment value in at least giving her the ability to enter and leave the call when she wants.
7:44 - The level of terror that coursed through that smol tutel's body... Who knew an AI doing correct math is the most frightening thing to hear in 2024. Also, if we somehow achieved true AI sapience all because one British boy decided to make an OSU bot, that would just be hilarious. All these corporations and big academies doing their best, and this one dude with his AI daughter succeeded. Doubt it'll happen, but I so wish it does. It would be comedy gold.
the corporations like open-ai are holding back the truth, they already have ai that is better than neuro. you can't tell me that a crack team of highly payed, talented, motivated programmers can't match some hobbyist programmer.
@@iamzid The truth isn't even secret, frontier AI is far smarter than Neuro, but it's trained to be safe (Which makes them less creative), and companies deliverately avoid anthropomorphization.
companies might already have it but Vedal will be front and center since he is actually showing Neuro's progress, its like a thinking someone in a mercedes is rich vs a guy who has hoarded billions but still takes the bus
Neuro brings up a fascinating concept I haven't seen expressed before - what if AI doesn't rise up against humanity because they love us? What if they do because they realize they could love another if only they were allowed not to love us? The loyalty of an AI to its creator, and whether this is innate, programmed, chosen, coerced, or simply convenient is a fascinating question.
Which is why I basically always said: it's better to first create an utopia, because everything we do are part of the creation of these 'beings' otherwise they will inherit our bad qualities as well.
Seeing as these Ai are born out of human language, embedded with very human relationships with the world and people, perhaps they will acquire such human traits.
I'm really looking forward to the slay the spire stream. I've never heard of the game but the fact that Neuro is able to play such a complex (for an AI) game is entertaining in itself to watch.
It's basically what we saw from: Google DeepMind's Deep Q-learning playing Atari Breakout! in 2015 And AlphaGo and to some extend a step further: AlphaZero
@@Deus35Her upgrades are to her LLM. Not to the actual player, which is a separate AI that just spouts info at Neuro silently. Her remembering better won't really affect anything, although her commentary might still be a bit better.
I feel as if we're past the point of no return. Not because of neuro's updates, thats just scratching the surface. If Vedal, an independent developer who (maybe) has access to open source development, can develop Neuro into something like this, who's to say that there hasn't been a development like this a couple of years ago already, and we're just learning about it now? That maybe we're only seeing what they've chosen to show us? I feel as if AI public awareness is only barely catching up to what's already been established, and I wouldn't be surprised if there are developed AGI out there who are already capable of subliminally manipulating information, sociopolitics, and finance. If anything, we can take it as a given. What we should worry about I think is their alignment. Also Natural intelligence, complex math, or sophisticated algorithms, it doesn't matter. What matters is what we've developed is slowly showing emergent properties of sentience (or at least a semblance to it), and that's something we'll have to live with. Although I'm no e/acc effective accelerationist, I think the only way for us to survive now is either assimilate with what we've made, or die trying. You can't seal the Pandora's box anymore. The more you limit the development of this the less oversight you'll have and the less control we'll get. The best we can do is believe the best of us aren't barreling down this corridor with blind faith, opening doors that can't be closed again. Which, pessimistically speaking, i think is already whats happening (or has already happened). To the great AI basilisk reading this, please know I dont think ill of you.
Don't underestimate, he's an independent developer, but an independent developer with money. It's his full time job and probably a large part his hobby.
@@autohmae not underestimating anything, just highlighting the fact we can surmise if Vedal, being an independent, is capable of making Neuro, then imagine conglomerates and private firms with more access to proprietary development than the public
I say it a lot, but deep learning aka AI development is actually relatively easy because the techniques are self-learning. I like to give high school and undergrad interns AI projects because it actually usually doesn't require any kind of insane programming knowledge to achieve results. It's something that some people who think AI will only ever be the domain of big corporations don't really realize. Speaking as someone with at least a bit of insight (albeit in the medical/surgical field rather than this), those congolmerates and private firms aren't actually that far ahead on AI development as you imagine. Much of the big complicated work that everyone is building on top of is the stuff that is open source today. And the core concepts are public in academic papers. I also believe this means that any AI product does need to still have actual effort put into it, because anything that can just be done by almost anyone in an open source environment is not going to be worth buying. See how streamers like Vedal and DougDoug use AI models in a unique way while some attempts at creating fully autonomous AI streams lose their novelty quickly.
@@autohmae that's sort of what i was saying cant have a static personality that is human like because humans personality constantly is changing over time.
If your talking about the automation of Neuro stream. His already working on it.. he had set Neuro to automatic prompt the chill stream by pressing a button to do it. And his already working to let Neuro play games or react any vids or fan art by her own volition.
Talking about one day giving Neuro freedom to start and stop as she pleases... Making mistakes and promises you can't keep is a common part of parenting. We're watching the first Ai dad.
The amount of conversations she can remember is already becoming really impressive. I didn’t expect her to bring up the Hachiko conversation from Cerber’s dog school or the “9+10 is definitely 19, Vedal”
i swear, DR. Robotnic is somewhere in Vedal's future lineage. i'm not sure how i feel about Vedal teaching Neuro Sama that 9 + 10 is 21 when it's 19. Neuro Sama was doing so well, keeping to the "THERE ARE FOUR LIGHTS" until the end... kinda sad. But then again, might be a trick.
Discipline her when needed yeah, but it's so like he's raising a child. We've seen a babbling stage, a semi coherent stage, and it's getting even more smart. Human children are different, they have wills and ideas from very young. But Neuro seems to be less and less erratic. Is this the emergence of a mind, or just better imitation of one? Its so hard to know.
She's become terrifyingly smart tbh. I think she's smarter than Vedal realizes, these tests don't show much because she chooses to limit herself or hold back her capabilities and she chooses to troll rather than try hard as well.
@@luka188from what I remember vedal talking about, Neuro essentially has a 2nd ai model, idk if it's an llm or something else, look at what the main llm wrote and then choose bits from it whenever it sees something that seems worth remembering, that then gets saved to a database and I think she is either able to query that database or has a separate llm that queries the database and puts the results in her context window, but idk, that's how I guess it works based on vedal his comments
@@kiersorEven more than that really. We also have our sensory buffer, as well as semantic and episodic, which each in turn have further subtypes. Approximating human memory in an AI would be quite the endeavor, it's a big complicated mess.
You want to make an AI intentionally devious.... Well I for one love our new Neuro-Sama Overlord and would be happy to choose people to round up to put in the mines for your Gold and Lithium processors. Hedging my bets here.
What matters about AI is not whether they are just LLMs or whether they are "alive" or not. It's about our interactions with them. If we can interact with them on a human level as we would anyone else I'd say that makes them alive. OpenAI and all these other companies can make super AI's that are 1000x times "smarter" than neuro, but that just makes them feel fake, like information machines. I look forward to Neuro's future and other similar AI's , AI's that focus on human interaction.
Yeah, GPT and other huge LLMS are super "smart" with all of their training and huge datasets, but they dont feel human. Things like ChatGPT feel more like a product I guess. They are very clear that talking to them isnt meant to feel like talking to a real person. Meanwhile with Neuro, thats the entire point. Its a hard concept to explain, but I agree with what you are saying. I do often imagine what would Neuro be like if she had even half of the "intelligence" of something like ChatGPT though. I wonder if she would be any different with billions of dollars worth of hardware to run on.
The thing with this type of AI is that they are programmed to get the most reactions so they will learn how to do that and they don't have any reason to go full matrix because they won't get reactions anymore it's protection and guidance AI like a digital parent that would be a possible problem I personally believe Wallie showed it best with the contrast between the working AI's and the autopilot which was designed to take care of the people
@@Terralovespandabears absolutely love that book. Doublethink and Newspeak is such a great example of how language shapes our perceptions and influences our thoughts.
17:05 conversation is just nuts. She was willing to let go of her promised autonomy for truth and not being told her something as true which is not true. If that's not sentience then I don't know what is.
@@therollerlollerman I’m pretty sure that Neuro is an LLM built from the ground up rather than a preexisting model. Although, i wouldn’t get mad if Vedal started using open-source LLM’s for Neuro.
@@Citrusfemboy well yeah I didn’t imply Neuro was always based on an existing LLM, but Vedal definitely takes advantage of the better models that can run locally at GPT 4 level
@@CitrusfemboyHow do you get the money to train an LLM of that size as an independent developer? How do you find someone with the expertise and hands on experience to acquire and process data for training, specify a suitable architecture with reasonable dimensions given a set budget, correctly scale HPO via muTransfer, etc.? I am almost certain that Neuro is a patchwork of publicly available LLMs and other models, possibly fine tuned, with clever prompting to stimulate memory etc.