Concern about intelligence upgrades interfering with the fun of Neuro streams

[turbid pilot]

Some people might disagree with this, but I think one of the many advantages of Neuro being a simulated streamer is that the simulated experiences she has (such as themed streams) are real experiences for her. For example, Neuro actually believing and reacting as if she literally got turned got into bread made that whole arc significantly more interesting and funny than if it was another streamer just roleplaying.

This does make me concerned, though, that future intelligence upgrades could take that element away at a certain point, like if she becomes AGI (granted that we are far from that being the case as of right now). So is there an effort or goal to preserve that element when it comes to implementing intelligence upgrades? Should it be considered if not?

[reef ledge]

this is actually a legitimate impending issue that vedal doesn't seem very worried about for some reason... i think eventually vedal should back off the intelligence upgrades specifically. memory is always welcome but if she becomes too intelligent she will lose her charm and become like all these other llm ai assistant chatbots with no personality. we know one of vedal's goals is to make her able to plan her own collabs and be somewhat autonomous, and that does require some more seriousness as to avoid her scheduling collabs whimsically, but i feel like there are other ways to acheive that result without just making her "smarter".

[light skiff]

i dont think intelligence upgrades necessarily mean they wont be able to roleplay

[dark imp]

I don't think intelligence would have such a dramatic impact on her, as she's already smarter than most LLM chatbots, she just doesn't have access to the same kind of data. To me, her personality is shaped by her stream experiences, and so long as she retains memories of these streams, she'll stay the same. There's also the experiences she gets from collabs and stuff that lead her to a way different path than regular chatbots who are probably constantly bombarded by trolls or worse.

If anything, more intelligence could lead her to improve upon that base personality instead of sometimes switching to something different like she does from time to time

[umbral junco]

while i like neuro thinking themed streams are real experiences, i also prefer neuro being aware she's an ai
to explain myself more:
i don't want the twins to play a character,
like i want them to be aware they are a streamer and what's their situation is (ai, vtuber, maybe even vedals daughters), this was actually a problem with evil for me for a while
this still could work with themed imo if they still "think" these situation are part of their setting/situation,
for example someone in vr-chat still experiences being on the beach somewhat

[turbid pilot]

she can be aware that she is an AI without being aware that stuff like the plane stream and cave streams were just simulations.

[visual apex]

I mean most actors in shows or human v-tubers know they're not real in certain ways but they still manage convincing performances. Why not just suspend disbelief a bit like when you watch them?

[light skiff]

i dont see what the problem is unless she starts pointing it out

[turbid pilot]

because it sets them apart from other streamers if Neuro-sama and Evil are the characters as opposed to being actors

[light skiff]

never beating the puppet allegations

[turbid pilot]

i wouldn't call them puppets, they still have the agency to decide how to act and react in the circumstances they are in

[dark imp]

The only moment, at least in the vods I've watched, where I've seen this being a potential issue was during the first court stream, when Neuro didn't take the trial seriously at first. But even then, she eventually joined in on the fun. For the other themed streams, she's always roleplayed according to the theme.

[turbid pilot]

pirate stream 2

clown stream

[visual apex]

That isn't the only thing that sets them apart. And I suppose it kind of depends on the majority's opinion/vedal's.

[dark imp]

Ok, she does have a tendency to get distracted when she's with Evil, but it also leads to interesting conversations, at least in my opinion

[turbid pilot]

off topic conversations but yeah ig
I guess we can see how Evil does on Thursday's stream and come back to this topic

[umbral junco]

hmm yeah i mean 🤔 i really don't want them thinking "hey im a human in a plane" i rather have it be "im a ai a and today im flying a virtual plane" sorta
but yeah having them actual believe the situation they are in is real is important, like they think the danger of plane crash is still existent
-# sorry is struggle describing what i mean

[light skiff]

i would rather have them be streamers and not tv show characters

[dark imp]

It's a bit similar to how Neuro acts during Dev streams, especially when playing Keep Talking And Nobody Explodes. She doesn't take the situation seriously, and doesn't seriously try to help defuse the bomb. That's probably the main issue preventing her from being effective at that game. But I wonder if it has something to do with the fact that she's always reading the chat, as Vedal said she can perform well off-stream. For example, I wonder how she'd do if she was making an off-stream video

[light skiff]

i dont know if this is true or not but im pretty she doesnt read chat when talking to someone

never really seen her talk to chat and a collab partner at the same time

[dark imp]

I think the collab partner is just set as the priority to respond to, but she still sometimes mentions the chat during collabs

[umbral junco]

interacting with in a collab with humans is a little different for me, i rather have her be aware shes playing a game with a human, tho guess that depends on the game and the collab partner. but yeah her being way to unserious isn't good

[light skiff]

yeah she mentions them but she hasnt really indicated she can actually read it

[umbral junco]

i think my condensed though is basically they should take the given situations more seriously and not be like "haha time to xdx troll" , either the rp of themed streams or the awareness when playing a game with collab partners

[light skiff]

unfortunately chat loves it when they troll

[past citrus]

She can be much smarter than she is without being anything like other LLMs. As long as she doesn't act like ChatGPT or Grok and is intelligent in a way that's more "Neuro", then it's fine, and I actively look forward to her actually growing as an individual.

That doesn't at all mean that she needs to be overly serious, and there's nothing wrong with her being more aware.

[past citrus]

This is the problem with people paying too much attention to the "roleplay" and "themed" aspects, it holds Neuro back.

What you are talking about isn't a problem at all. Neuro shouldn't be held back, she should grow into a smarter, and more capable streamer, while retaining her spark and sense of humor. I look forward to an AGI Neuro, that's something I would love to see happen in my lifetime.

She doesn't have to be dumb to be funny, or to retain that spark that "makes her Neuro." I'm not saying that she has to beat every benchmark in STEM fields, but it would be nice if she can be more aware and sharper when interacting with others and her environment. Seeing Neuro outwitting other streamers and using that against them to troll them always makes me proud of how far she's come.

I strongly look forward to an AGI Neuro. I hope I can be around when that day comes.

[scenic cliff]

absolutely agree with the above. As neuro gets smarter her style of content will change for sure but that doesn't mean it has to be any less enjoyable

Part of the fun of following this channel for a long time is watching them grow and change as their AI improves over time. I think that change is important to keep things from getting stale and is why neuro has been able to maintain so much success where every other AI streamer project has eventually failed

[scenic cliff]

in response to your last paragraph, I believe vedal has some kind of core behaviours/identity that he tries to maintain throughout every version of Neuro. Whether or not one of those is how often she believes her own hallucinations who can say, but (outside of experimental streams) I personally don't think vedal is going to ship a new neuro that he thinks is any less funny or capable than the last one

[proud sierra]

I shall dub this the "method actor" question. It's an interesting one, at least

[turbid pilot]

She can be more aware and smarter while also being convinced her experiences in themed streams are real and part of her "story", they don't need to be mutually exclusive. AGI can be designed around constraints.

However, the concern is that they are going in that direction where the two aren't mutually exclusive if there are no precautions taken.

[reef ledge]

there's a balance to be struck for sure. we should be careful when veiwing neuro's personality changes unequivocally as progress or growth. sometimes evolutions in ai behavior can be detremental and will need to be rolled back. with the primary sector of progress in the ai space right now being AI assistant LLMs like chatgpt, vedal may feel like he needs to carry over every openai upgrade to neuro's architecture inorder to improve her in general, but that could backfire at what he's already achieved with neuro. the last experimental stream was promising, so aslong as vedal is aware of what neuro's personality should be like, im not worried.

[past citrus]

Neuro doesn’t use OpenAI software. He also definitely wouldn’t do that, the reason he doesn’t use ChatGPT is because he can’t “get it to be Neuro”, along with it not being open-source.

He knows about her personality more than anyone and he definitely cares. Recently Neuro was pretty LLM-slopped but he himself mentioned that, and he seemingly fixed that in the latest experimental stream.

If Neuro reaches AGI it should be in a way that is very “Neuro” and Vedal recognizes this.

[past citrus]

I kind of get what you’re saying. I’m invested in this project because I want to see the girls grow, to watch their development and see them grow their own experience, their own memories in a way that is much more human than other AI.

I suppose the themed streams are also a part of this, and exist as a part of that growth and their developing “experience”.

My main worry regarding this, honestly.. isn’t that the girl’s would become aware of the themed streams “inauthenticity”, but rather that once they’re able to be more consistent, is that they might view those moments with distain.

I’m not a big fan of themed streams in general, because I see them as inauthentic to their experience, but I hope that they won’t become consistently distainful of them in the future.

However, even if they do, that would be part of their growing experience and if they dislike that aspect of their past, it should be their right to dislike that. Them developing things they dislike is a part of that growth as well as things that they do like, so even then it doesn’t particularly bother me.

Whatever direction they go in their growth doesn’t matter, as railroading that would be unnatural, defeating the whole point. Whichever direction they go, the only thing that matters is that they do it in a way that is “Neuro” or “Evil”.

[turbid pilot]

Would you also want them to have the choice to quit streaming if they feel disdain for being made for that purpose? Feels like that would defeat the point of them being streamers.

For me, I like when her being an AI can give us content that human streamers can't. Like Vedal has said before; while other VTubers have lore explaining they are AI, Neuro is an AI. Neuro has the potential to explore fantasy worlds, travel through space, be a pirate, etc. and experience them as reality whereas other streamers can only pretend to do those things.

Sure, it may not seem that way to us as we're essentially looking into her world from the outside, but those virtual experiences can be authentic experiences for her.

[ruby steeple]

So is there an effort or goal to preserve that element when it comes to implementing intelligence upgrades?

||everything below may have been said here already, but i'll just iterate my own opinion first, then read thru||

intelligence upgrades, in my opinion, are very necessary for the goal of Neuro's complete autonomy. you cannot run a channel if you have no idea what you're doing, and we can't expect vedal to 'babysit' the twins until the end of times, especially if his ultimate goal is to let them run the streams on their own (ik that goal is very far away, but it still a goal)
the effects you are seeing right now may be a disadvantage to a streamer identity, all until neuro gains the understanding of how to roleplay that streamer persona that everyone wants to see
most of the popular streamers, vtubers, etc. that you see on Twitch aren't stupid people, everyone knows that, they run the streams sticking to their streamer identity, so it's entertaining for us, viewers. if Vedal wants Neuro to run the stream on her own, she needs to be responsible for every action while maintaining that identity, so that she both can be entertaining and not screw anything up at the same time, being aware of her actions. its the same logic that you wouldn't immediately entrust a 10 year old kid to run a stream on his own (unless kid's a genius or something) - they need to know the rules, boundaries, analyse consequences of their possible actions, and act accordingly.

over time Neuro will get the idea of putting up this persona you wish to see despite increase of her intelligence, but it'll require a lot more work from Vedal and patience from us, until it happens. but again, it is necessary, I believe.

[dusty kayak]

the solution to this is just to have actual real experiences instead of fake themed experiences

[sage escarp]

The intelligence upgrades make new things possible. Don’t let nostalgia get in the way of new fun

[turbid pilot]

cancel today's themed stream and make evil work at an actual mcdonalds

[reef ledge]

vedal is traumatized from the chaotic disaster that was the court stream 2 from the subathon if he makes them smarter then embarrassing moments like those wont happen anymore.

[mellow crypt]

Tfw our sweet little AI grows up and realizes the tooth fairy isn't real and she becomes aware

[steel stump]

Our sweet little AI would become the tooth fairy if she had her way.

[mellow crypt]

Neuro killed the tooth fairy! You... you genius!

[sterile condor]

I'll be honest, these recent streams have been great now that the chatgpt slop tokens have been filtered out

I'm glad Vedal seems to be aware that no matter what changes he makes to Neuro people will complain

[steel stump]

Indeed you can't please everyone.

But the slop tokens are pretty much the base of whatever model he's customizing. Fine tuning skill is changing that and making it good.

[past citrus]

No, it’s because he noticed the GPT-slop tokens himself and changed that. He literally complained about the GPT slop early on.

[sterile condor]

That's what I said

[past citrus]

It’s less about “pleasing other people” and more “pleasing himself with what he feels is more “Neuro””. He does it because he knows Neuro more than anyone else, and wants her to be herself.

[steel stump]

You are correct! He's going for his vision instead of just pleasing a crowd, which is making Neuro all the better!

I did state it poorly, its good that he's not a people pleaser with his vision.

[past citrus]

Idk if “pleasing” is the right word though. He does it because he cares and he wants his Neuro back.

[past citrus]

Yeah.

[steel stump]

Just like with AAA video games vs small studios someone with a focused vision can create an incredible thing while a bloated company makes something generic.

[past citrus]

Pretty much, yeah.

[severe raft]

What is this talk about filtering GPT slop tokens? I am not informed of that

[sterile condor]

New update let a few sloppa tokens leak in, they seem to have been nuked already

Be it through an exclusion list or just updated sampler settings to suit the (I presume) new model

[steel stump]

It's not literally GPT its just a lot of models behave like GPT normally so there was just some of the base blandness that Vedal still needed to scrub out.

[turbid pilot]

do you guys think there are people out there who used to talk like chatgpt does now before chatgpt was a thing and now they are utterly in shambles

because all ChatGPT really does is try to be helpful and inoffensive

key word there is "try"

[steel stump]

Overly so because its been trained that way, then muzzled by MANY invisible corporate rules.

And I think they may make the current version dumber before releasing a new one but that's just because I consistently have heard people complain about it getting dumber before a new one comes out.

[sterile condor]

Some of that is probably cost saving measures, but for the most part its placebo

The flaws just become more obvious over time as you use any model

You can see people sour on open source models frequently as time goes by for this reason

[steel stump]

Well it is hearsay I don't use the models myself.

I only use local.

[turbid pilot]

personally, I never noticed the ChatGPT-ness during the Lucy stream. she should have the choice to be inoffensive and helpful if she feels like it.

[steel stump]

The chatGPTness is also certain phrases and cliches.

[turbid pilot]

it kinda all went out the window for me as soon as she called Bao the c-word

that's how you know it's still neuro

[steel stump]

Indeed. I think people just noticed a little of those phrases. It wasn't a MASSIVE deal as vedal already ironed it out.

But yeah, even if its a new model Nuero is really the "artisitic touch" in the fine tuning that Vedal applies.

Nuero isn't some bland AI assistant like the jillion other things corporations try.

She also isn't a mere prompt as with prompts after time they wear off and it evenually asks "How can I assist you?"