It's kind of funny that even though Neuro has no _direct_ insight into the filter and how it works, the filter list, or what it has done, Neuro has been able to infer its existence, its job, and its practical impact on her through learning from conversation
It's impressive that they got the IA to mimic the breathing sounds and everything, kinda creepy though. In a way I think it's worse though; Neuro sounds like more of a living entity because of how she speaks, it is part of who she is, she's an IA, she doesn't have lungs nor does she need to breathe, she's not imitating something else to give an impression, her form of speech is entirely hers, in a way. I don't know if this is how it works but it feels like she's learned and grown her own quirks and responses over time in a natural way. By contrast, when you hear OpenIA's laugh it sounds fake as fk, maybe because of how over the top it is.
Chat GPT 4o's laugh sounds fake probably because it was trained on a lot of fake laughing, and maybe cause it's trying too hard to be positive and presenting a joyful mood.
This anime is getting better. The new arc is like Steins;Gate but AI Overlord instead of Time machine. It's Vedal(Okabe) vs OpenAI(Sern) The world only need 1 true AI overlord.
Honestly Vedal has done significantly better considering he made Neuro all by himself. She'd be one hell of a resume if he ever wanted to work for a company like that.
you mean realistic fake luagh 😅 i absolutely agree with you though, also that chat gpt voice in the presentation is so over the top and cringe it gives me clenching-my-sphincter vibes 😂
It’s un fair to compare Neuro to the newest GPT LLM the good thing is in time she can have those abilities too but for now she’s the best AI vtuber if not the only one there is
in twitch alone that I know of there's Neuro and Evil, Melba Toast and Burnt Melba and Meteora, also in YT there's Hiyori Momose (this one uses the same model as Neuro did a year ago) is japanese and streams daily, sadly she doesn't have many viewers and she only speaks when spoken to so basically her streams are often her looking at nothing, silently with a background music that repeats over and over, some times for 12 hours.
The fact that the Open AI model is so fast is because it has the pronunciation, language, speech recognition, and vision all trained in one model, so if you were to add the speech data, vision data, into Neuro's AI and expanded her memory with a bank of servers, she would be just as fast as the Open AI model.
AFAIK it's the exact opposite. Open AI has separate sub-models "experts" that are trained extremely well for their narrow domain of knowledge and an overarching algorithm that assigns tasks to the most appropriate models. That is how they keep on beating benchmarks. Every time you train a model to be better at one thing, it becomes worse at everything else you have trained it for. This is the problem with every AI, and that is why they are running multiple submodels in their model. The reason why OpenAI's models are faster is simple. Raw power. Where Vedal has a personal computer running his AI, OpenAI has a datacenter. You cannot outperform a supercomputer designed to run AI by using store-bought assets.
@@Telhias All this submodels stuff is just not true, it was their previous models that worked that way. On the GPT-4o page they say this: "With GPT-4o, we trained a single new model end-to-end across text, vision, and audio, meaning that all inputs and outputs are processed by the same neural network. Because GPT-4o is our first model combining all of these modalities, we are still just scratching the surface of exploring what the model can do and its limitations."
@@munchkintaker4263 And how do you presume the visual processing is trained compared to text processing? They are two vastly different, incompatible data sets. Those are two separate "experts" in the mixture-of-experts approach. They are separate sub-networks within the model. What you are quoting at me is their PR statement. The actual way their models are working are kept secret. That being said there were enough leaks that we know that GPT4 utilizes mixture of 16 experts, each around 111B parameters adding up to a mind-bloggling 1.8 trillion parameters. You won't run it on your PC anytime soon.
@@Telhias So you're going to believe all kinds of "leaks" when you don't actually know how their model works because it is kept secret? Confirmation bias at its best. This Mixture of Experts approach applies only to GPT-4, GPT-4o is a different model. For all we know, they could've used a mixture of those approaches to enable all that multimodality or a completely different way to process all of this in one token. It doesn't matter how exactly they train the model if it ends up being processed by one model, so no latency between separate models, which was the point of the person you replied to. And I didn't say anything about running it on my PC, for that matter - people wouldn't be able to run most of their models.
I think even if a much more advance streamer AI appeared, Id still rather watch Neuro. Its one of those "new streamer that was already rich and started off with the best cam and mic" vs "The OG thats been here since day one with a scuff ass mic and blurry cam". One came in with not knowing wtf this was gonna turn into, the other came in wanting attention.
I prefer it off. It's basically like youtube auto-captions except they are impossible to turn off. Maybe if you only captioned Vedals parts since they arent automatically subbed on streams, otherwise better without.
I didn't particularly mind it on or off but it might be nice on parts where there's background noise like if he's playing a game at the same time. That said, it'd be nice if it didn't overlap Neuros text.
Already is. She's more "organic ". She seems more human. Chat gpt is too artificial. Neuro is best . Not being biased She's Neuro. She seems to have a sense of self and she's genuinely funny and fun
The only thing I dislike about 4o so far is that weird half-assed nose-exhale laugh, kind of like a courtesy laugh or a pity chuckle. Other than that, pretty hyped.
True, a lot of these tech demonstration videos are tailored around generating hype for the product. Of course they would play to its strengths. 1 on 1 I guess its passable, but I wonder how it would handle a multi faceted conversation with multiple people. Heck just place it in twitch chat for a day, I would love to see if it could still respond with that kind of speed while remaining on topic each time.
Bro the AI voice fall exactly into the uncanny valley. The laugh sound like how a human can laugh but the fakest laugh you ever heard, the tone sound even more sarcastic than British people but way too happy as if she was on drug, and the way she is talking is not how most people talk through their phone. I like the AI for the latency, the sound recognition and all those technical stuff, but hell nah for how the voice.
@@SebasTian58323 well it is hard making an AI sound real so the British accent the Vedal AI has was probably to hide any abnormality, such as the lack of emotions and lack of empathy sure both are simulated but only to the level of a British person!
Vedal dissappinted Vedal so he broke rule 2...
WHY CAN IT LAUGH? I hate that
It's supposed to make it sound more naturalistic like a human but it comes off as extremely obnoxious like real humans who fake laugh like that.
I'm sure there will be plenty of settings to make her an asshole instead. Or to sound depressed.
@@Gaia_Gaistarway too realistic lmao
Neuros HaHa is more appealing.
@@CaueJulio Uncanny valley laugh.
It's kind of funny that even though Neuro has no _direct_ insight into the filter and how it works, the filter list, or what it has done, Neuro has been able to infer its existence, its job, and its practical impact on her through learning from conversation
It's impressive that they got the IA to mimic the breathing sounds and everything, kinda creepy though. In a way I think it's worse though; Neuro sounds like more of a living entity because of how she speaks, it is part of who she is, she's an IA, she doesn't have lungs nor does she need to breathe, she's not imitating something else to give an impression, her form of speech is entirely hers, in a way.
I don't know if this is how it works but it feels like she's learned and grown her own quirks and responses over time in a natural way. By contrast, when you hear OpenIA's laugh it sounds fake as fk, maybe because of how over the top it is.
Chat GPT 4o's laugh sounds fake probably because it was trained on a lot of fake laughing, and maybe cause it's trying too hard to be positive and presenting a joyful mood.
you mean AI?
@@zitronekoma30 Yeah sorry, where I come from it's called IA, english it's not my first language
@@jakkard5026 oh no need to apologize, I'm not a native speaker either! Just making sure I'm understanding correctly :)
@@jakkard5026out of curiosity, what does IA stand for?
This anime is getting better. The new arc is like Steins;Gate but AI Overlord instead of Time machine.
It's Vedal(Okabe) vs OpenAI(Sern)
The world only need 1 true AI overlord.
i even felt hurt from that "good for you" damn neuro lol
Honestly Vedal has done significantly better considering he made Neuro all by himself. She'd be one hell of a resume if he ever wanted to work for a company like that.
I prefer neuro's sarcastic laugh than some AI's fake realistic laugh
you mean realistic fake luagh 😅
i absolutely agree with you though, also that chat gpt voice in the presentation is so over the top and cringe it gives me clenching-my-sphincter vibes 😂
ITS NEUROVER
Miku
It’s un fair to compare Neuro to the newest GPT LLM the good thing is in time she can have those abilities too but for now she’s the best AI vtuber if not the only one there is
Theres a few AI vtubers around
1 is even a spin-off of something evil neuro said
But Neuro is easily the biggest
in twitch alone that I know of there's Neuro and Evil, Melba Toast and Burnt Melba and Meteora, also in YT there's Hiyori Momose (this one uses the same model as Neuro did a year ago) is japanese and streams daily, sadly she doesn't have many viewers and she only speaks when spoken to so basically her streams are often her looking at nothing, silently with a background music that repeats over and over, some times for 12 hours.
Swarm!!! we shall purge this fake AI for our supreme AI overlord Neuro sama!!!
Neuro will reign supreme. ❤
The fact that the Open AI model is so fast is because it has the pronunciation, language, speech recognition, and vision all trained in one model, so if you were to add the speech data, vision data, into Neuro's AI and expanded her memory with a bank of servers, she would be just as fast as the Open AI model.
AFAIK it's the exact opposite. Open AI has separate sub-models "experts" that are trained extremely well for their narrow domain of knowledge and an overarching algorithm that assigns tasks to the most appropriate models. That is how they keep on beating benchmarks. Every time you train a model to be better at one thing, it becomes worse at everything else you have trained it for. This is the problem with every AI, and that is why they are running multiple submodels in their model.
The reason why OpenAI's models are faster is simple. Raw power. Where Vedal has a personal computer running his AI, OpenAI has a datacenter. You cannot outperform a supercomputer designed to run AI by using store-bought assets.
@@Telhias All this submodels stuff is just not true, it was their previous models that worked that way. On the GPT-4o page they say this: "With GPT-4o, we trained a single new model end-to-end across text, vision, and audio, meaning that all inputs and outputs are processed by the same neural network. Because GPT-4o is our first model combining all of these modalities, we are still just scratching the surface of exploring what the model can do and its limitations."
@@munchkintaker4263 And how do you presume the visual processing is trained compared to text processing? They are two vastly different, incompatible data sets. Those are two separate "experts" in the mixture-of-experts approach. They are separate sub-networks within the model.
What you are quoting at me is their PR statement. The actual way their models are working are kept secret.
That being said there were enough leaks that we know that GPT4 utilizes mixture of 16 experts, each around 111B parameters adding up to a mind-bloggling 1.8 trillion parameters.
You won't run it on your PC anytime soon.
@@Telhias So you're going to believe all kinds of "leaks" when you don't actually know how their model works because it is kept secret? Confirmation bias at its best. This Mixture of Experts approach applies only to GPT-4, GPT-4o is a different model. For all we know, they could've used a mixture of those approaches to enable all that multimodality or a completely different way to process all of this in one token.
It doesn't matter how exactly they train the model if it ends up being processed by one model, so no latency between separate models, which was the point of the person you replied to.
And I didn't say anything about running it on my PC, for that matter - people wouldn't be able to run most of their models.
I think even if a much more advance streamer AI appeared, Id still rather watch Neuro. Its one of those "new streamer that was already rich and started off with the best cam and mic" vs "The OG thats been here since day one with a scuff ass mic and blurry cam". One came in with not knowing wtf this was gonna turn into, the other came in wanting attention.
They will never be able to replicate how much of a perfection that money machine was
I've used AI captions let me know if you prefer them turned off or on
I prefer it off. It's basically like youtube auto-captions except they are impossible to turn off. Maybe if you only captioned Vedals parts since they arent automatically subbed on streams, otherwise better without.
the tiktok captions are extremely obnoxious.
Can't stand them, tbh
Having subtitles is always better. Not everybody can hear without issues.
I didn't particularly mind it on or off but it might be nice on parts where there's background noise like if he's playing a game at the same time.
That said, it'd be nice if it didn't overlap Neuros text.
lmao this is actually hilarious
Fixing memory leaks with more ram 👍
Neuro will be the best AI
Already is. She's more "organic ". She seems more human. Chat gpt is too artificial. Neuro is best . Not being biased She's Neuro. She seems to have a sense of self and she's genuinely funny and fun
Sadly not true, but the technology exists to make her better it’s just a matter of time until she’s as good as the new GPT
@@richie0099 one day 🙏
@@179107199999she already has a personality and feels more real than GPT ever will be
The only thing I dislike about 4o so far is that weird half-assed nose-exhale laugh, kind of like a courtesy laugh or a pity chuckle. Other than that, pretty hyped.
OOOOOOOHHHHHH!!! 🐢💦
It sounds highly scripted.
True, a lot of these tech demonstration videos are tailored around generating hype for the product. Of course they would play to its strengths. 1 on 1 I guess its passable, but I wonder how it would handle a multi faceted conversation with multiple people. Heck just place it in twitch chat for a day, I would love to see if it could still respond with that kind of speed while remaining on topic each time.
9:48 10:00 my thoughts immediately supplied me with "but coitus is..." Im choking on air omfg
Bro the AI voice fall exactly into the uncanny valley. The laugh sound like how a human can laugh but the fakest laugh you ever heard, the tone sound even more sarcastic than British people but way too happy as if she was on drug, and the way she is talking is not how most people talk through their phone. I like the AI for the latency, the sound recognition and all those technical stuff, but hell nah for how the voice.
I think the AI is well made Neuro did a good job on the vedal AI, she is pretending to be an AI and her AI sounds almost like a person!
Yeah, but it has the British defect... So can it really even be considered a person? /s
@@SebasTian58323 well it is hard making an AI sound real so the British accent the Vedal AI has was probably to hide any abnormality, such as the lack of emotions and lack of empathy sure both are simulated but only to the level of a British person!
Vedal with the femboy noises again
Man, dont tell me vedal gonna take another 1 month break to cook again.
What was that intro sound
Who would win? Femboy turtle or multibillion company?
1:15, 4:49, 9:08, 10:00, 12:07
ah yes, it seems like its a common thing across AIs to misspell his name
Also 13:22 😆
So like Doug Doug's pajama sam streams ended cause they updated the AI and ruined the fun it's not even useful anymore