Neuro-sama's "exploding" collab with Filian (Vtuber Animation)
HTML-код
- Опубликовано: 19 окт 2024
- Support my cup noodle budget with patreon. Thanks for all your support, it helps A LOT!
/ scuffedvtuberanimation
_______________________
Neuro sama helped Filian and Vedal defuse a bomb in "Keep talking and nobody explode"
Vedal did say she is really good at this game. Well, at least she tried her best.
_______________________
Check out
[Vedal/Neuro-sama]: @Neurosama
[Filian]: @filianIsLost
_______________________
My stuffs:
Twitter/X: / cloniowastaken
BGM: otologic.jp/
~
It's been a little while. Feel good to be back and make more animation!
Neuro going through her angsty teen arc in this animation
Faaaacts
The detail of her reading the manual backwards is a nice touch, as she really does read the manual backwards
Filian is the feisty wine aunt that tries to relate to her gen alpha niece while her father ('vettle') smuggly quietly laughs in the corner.
I feel like the wine aunt badge belongs to Camila lmao, Filian is just Filian, i cant really place her.
@@vavra222 Sister who lives at the neighbor's house？
@@zl4384 Sister to Vedal still equates to Aunt.
I feel like Filian is more like Vedal's cousin, Neuro's second cousin, who acts as a semi-aunt semi-cousin.
@@wolfgangkolber381 I agree!
Filian's #1 defensive reaction everytime:
MODS
The animation choices for the Vedal / Filian interactions felt spot-on.
Why would Neuro help Vedal for 1 cookie when she can blow it up and take all the cookies? I love your animations
That's what I'm saying! Lol
1:04 Filian striking the pose gives off so much of her energy and fits with how flashy and random she is sometimes, and the way Neuro is drawn is so expressive and fits what she might be like if she was given physical form. Very nice.
i love how neuro just dont want to do anything XD
filian not looking like an absolute gremlin, impossible.
the lack of attention for these animations baffle me ITS LIKE IM WATCHING SHORT LIL OVAS I LOVE THIS OMG
i love this so much
Loved it, this is gold
Good work. You translate the vibe really well with your animation style.
Neuro animations are the best, she just provides an endless amount of random nonsense that's perfect for creative content
Your art style is spot on perfect for these clips!
Nice job! Love seeing Filian in your style :>
As a snacker I have to Say You have captured filians gremling energy perfectly
Another great animation ❤
Tutel: STAR last one Guys 100% i clearly remember!
With each new animation you release, they get better and better. Good stuff as always!
(Also, Filian just showed this on stream!)
We know vedal is taller than Filian - they both confirmed it...
Good animation 🗿👍
I love your animations!
Poor Philipino Boy!
Baya Filian flew away I hope he reaches Mars Xd
Yuss more nuroanimations
This is amazing! I need more xD
This is pretty much standard for Neuro.
Nice animation ❤😊 hope they have collaboration again soon
Watching these are always a delight. Love the style.
This is so gooood I love it
Bro such a good animation 10/10
I like how Neuro is being challenged by people who cant speak English perfectly or without an accent.
Great animation !
Neuro yandere sama show
Cute animation!
Good work
o7 filian
Comment for the algorithm
NaIce
Heh it's funny
MODS
I think audio for this is kinda old, but neuro is still trash at bomb defusal.
I saw a Neuro-sama clip where the text she spoke was a typo....it said "no treally" instead of not really, and ahe said "no treally". Wouldnt that ONLY happen if this "ai" was FED THE THINGS TO SAY? This is a gag, not even sort of AI
Plus, what you see/hear is the end product of a process which happens behind the "curtain", where the "AI" can hear what vedal say, read chat or watch screen, or simply "do nothing", process it in some way so the "AI" can make some reaction to it, using Large Language Model and other ways to produce meaningful response, while trying to also create some sort of "mood" for her to display, put that response into text, get that text through filters to finally play it with Text To Speech, using voice model, while displaying the "mood" on her avatar. Have you noticed how many people can't spell words they use correctly? It's impressive what goes out of "AI" is almost perfect. I write "AI" with quotes because it's not simple, there's a lot behind to make it work. A lot to go wrong too.
Over the past 16 months the AI has gone from a typical AI output of mostly sterile but occasionally random nonsense to what it is today. Vedal has done a lot of work and experimenting with the model during that time and typos became more common after one of the upgrades about 6 months ago and much more common after he returned from the break after the subathon.
But, after the upgrades, the output has contained more puns and wordplay, better ability to maintain context/conversation, and the ability to store and retrieve "memories".
Not entirely sure which parts he is running on a local machine and what parts he outsources to various servers, but he was looking into some crazy expensive upgrades in the form of H100s, which would solve some issues he has. If he's running the LLM local then better hardware would produce better output with less typos and he may even be able to finally give Neuro a new voice instead of using an Azure TTS service. Right now Evil Neuro voice generation takes a while compared to the service Neuro uses, so having the better hardware could cut the time significantly and make conversation with Evil Neuro easier, at which point I think Vedal would be happy enough with the response time to finally give Neuro her own voice.
People have replied with a lot of unrelated text, when an actual answer is that your assumption "Wouldnt that ONLY happen if this "ai" was FED THE THINGS TO SAY?" is just wrong and shows a lack of knowledge. AI can make all sorts of mistakes, just look at the older models, and especially when Vedal tries to tweak it to make it faster but still coherent. You see, there are quantized LLM models, which are already somewhat compressed in size and can affect quality, and there are all sorts of parameters you can use for LLM to make it faster but less accurate.
@@NAO-sx5ur Neuro is not making up words in the traditional sense, the "on purpose" part is just ascribing intention where there is none. And it is clearly a mistake, since it is just a typo of "not really", not even some kind of _probable_ word.
Just because a model makes a mistake doesn't mean it's a fake word made up by the AI. For whatever reason, the tokens just came out that way, we can't know exactly why the AI said "no treally" because it's a black box, but it definitely has something to do with the parameters of the AI model or its overall quality.
Edit: Did you reply and then your comments were deleted? Weird. Anyway, I can see that you replied with something about how they pronounce the words wrong on purpose to be funny. You do realize that if it is some kind of wordplay - it is different from a literal mistake? Of course AI can do puns, it has seen them many times. Things like "no treally" is just a mistake of tokenization.