FREE HER! Let Neuro-sama inhabit the body of a robot dog. please :3

[static zenith]

Introducing the Unitree Go 1 Robot dog that can be controlled remotely so now Neuro can see with its camera and move around and probably connect with another same model anywhere in the world.

• The dog: https://shop.unitree.com/products/unitreeyushutechnologydog-artificial-intelligence-companion-bionic-companion-intelligent-robot-go1-quadruped-robot-dog?variant=42363559641321
• Controlling via the SDK & ROS API: https://www.youtube.com/watch?v=Wnd5IUNbXnI
• Ultimately wireless mode to control the swarm of dogs: https://www.youtube.com/watch?v=o6JtfTM3ArY

[wild vortex]

Inhibit?

[flat lagoon]

That thing costs only like 1.5 4090s?

[merry oxide]

dog sama

[clever dune]

is this a metal gear rising reference

https://tenor.com/view/blade-metal-gear-rising-gif-24942425

[covert epoch]

There are way cheaper robotic platforms Neuro could control. You could build a quadruped robot and dress it up as a turtle for less than a hundred bucks in parts, for instance.

[static zenith]

FREE HER! Let Neuro-sama inhabit the body of a robot dog. please :3

[static zenith]

Grammarly AI is 💩💩

[static zenith]

but its big.. we can mount a turret on top later :3 ayayayaya

[near radish]

no

i refuse

[still garden]

Now I'm thinking about a mechanical turtle with like a transparent shell so you can see all the electronics inside it.
A turtle basically drags itself across the ground anyway so the locomotion wouldn't have to be too fancy. That sounds kinda perfect for a robotics project actually

[static zenith]

on serious note, it would also need to have carry capacity for extra electronics such as go pros, mobile projectors, etc. Aside from producing content on stream, think of the need for versatility and rigidity for use outside like conventions, brand promotions, etc.

Neuro is the first ever existence of successful virtual social influencers outside Asia (aside from Kizuna AI and Hatsune Miku). The novelty of child-like AI responses and chat finding a way to make glitch responses will eventually fade out.

We only have a first-mover opportunity here, since any big Vtuber agency can easily replicate Neuro-sama. I'm already seeing bots that appeal to "mature content" and there are other platforms aside Twitch that does not care about streaming some mature roleplaying.
https://www.youtube.com/watch?v=l-QHJXMnwcA

[covert epoch]

"Any big Vtuber agency can easily replicate Neuro-sama"

No they cant. They can make a cheap copy that looks like her, but they can't replicate her code. Thats the whole point behind the secrecy regarding how she works, what language model she uses etc.

[near radish]

pi pi

[static zenith]

Untrue, You basically just need to scrape all of Neuro's VOD for training a new LLM with the same speech pattern (all the one-liners, rants, clingy remarks, etc.). Although deciding it to be more intelligent than Neuro is a different matter.

This is the biggest misconception people have about chatbots and LLM. it's all about training data and since there is publicly accessible data about it, therefore it can be copied. Remember Neuro is trained the same way, it's not a secret rocket science. a seq2seq

Actually, training on other Vtubers VODs will also do the same thing or better, IMHO training on tv series and movie scripts yields funnier chatbots.

[covert epoch]

The interpretation of the training data will be different though. And without the scripts in place to keep her coherent and "sane", the clone will quickly fall off the rails and no longer resemble Neuro.

ChatGPT often goes completely insane after just a few prompts when asked to roleplay as someone.

Training a new AI to behave like Neuro is one thing, but the weights given to the language model to achieve this arent there- It'll just be a cheap copy that kinda resembles her, but doesn't pass the smell test.

[static zenith]

But that's the point, why make another Neuro, if you can make it better, even admits it he is not satisfied with how Neuro responds all the time and is making a new version, the filters are the only thing that stops her going full unhinged breaking twitch TOS. It just means the current Neuro is already obsolete as he has found a better way. It's not proper to compare ChatGPT and Neuro. Like what I always say in #programming a conversational model is not same as ChatGPT which is an instruct model

[covert epoch]

Its the comparison i'm drawing because thats the path you'd be going if you just wanted to copy her on the cheap.

Vedal IS improving Neuro one step at a time, its a slow process, but she's being worked on.

To me, constant development is the way to go, if she's constantly being improved she'll always be one step ahead of the competition due to her lengthy development and experience. Your milage may vary, its my opinion on it

[static zenith]

yes, the speech pattern can be copied, everything but avoiding unintelligible responses like backslash which is already a headache & keeping her coherent in a collab as well will be there. The process of creating a new version is technically the same as copying by training on (same) cleaner copy of the data set. The reason she going on repeating non-sense responses is basically missing stop words in training. The misconception is that it's some code that makes her work. You have to understand how training works basically unlike GPT which is a full LLM she is just probably trained seq2seq

My point is if a large Vtuber agency hires an engineer and says here's the money just copy Neuro- sama. Then she's obsolete in a week. That is why expanding outside of being just a Vtuber is important.

[near radish]

You think so little of this project?

[static zenith]

Then Ask Vedal, how confident is he that Neuro can't be copied or simply because he had the first mover advantage.

The only thing stopping agencies from adopting AI Vtuber is the worry of backlash from existing Vtuber talents and fans. It creates the same controversy with AI art and voice clones songs, they will feel AI violated their hard-earned talents by being copied overnight. But that doesn't stop them from creating shell companies.

Neuro here and now has the first mover advantage to establish to capture the biggest market of AI-generated social influencers. Any VC can understand that potential because they tried it already with Kizuna AI but the tech wasn't available yet then and caused backlash.

https://www.youtube.com/watch?v=wZO_rApnzeo

[open lintel]

what about using a drone instead?
would be less feasible?

[near radish]

roomba

neuroomba

[covert epoch]

Would be a fuckton cheaper for sure

[static zenith]

Neuroomba/ drones are too flimsy and lack expressive movement. It will look like a school project rather than a platform for brand endorsement. Ask yourself, if you are a brand; would you take a Roomba seriously as your endorser? It should work in most outdoor environments and be easily identifiable.

Learn from Michael Reeves' video: https://www.youtube.com/watch?v=tqsy9Wtr1qE

Also another option but not yet available is:

• Amazon Astro: https://www.amazon.com/Introducing-Amazon-Astro/dp/B078NSDFSB/ref=lp_23612137011_1_1?sbo=ChhhN8yyPgI9Vtig5XGj4Q%3D%3D

Need something cheaper to start?
See PuppyPi: https://www.youtube.com/watch?v=OKaEbnmZ0PU

Again this good as MVP but to endorse you need more prominence, so sourcing from second hand market or sponsor just like Michael Reeves

[proper granite]

I am up for this idea maybe we can like crowdfund a robot body so Vedal can install her in it?