#ABANDON THE ARCHIVE!!

1 messages · Page 1 of 1 (latest)

wanton venture
#

So, I got to thinking , after the Reptile Lord showed us the official corporate harassment notices for taking too long, what if he...

  • Has Neuro and Evil write the script for an entirely new game (given their different models and how they've come to see the world, they could each make rather unique stories)
  • Have each of them do one of the branches of the story (so at least 2 paths, but additional ones possible and could even intermingle the events, like those True End things, the ones that only make sense, once you've played from the other characters' perspectives)
  • Set it in the same world as Abandoned Archive
  • Current AA systems could still be used as part of the gameplay mechanic, but it would be more like just the events and dungeons, not the main game concept

This also makes it look good for Steam, because it can still be released on that page, and the demo is more like just the dungeon minigame fighting mechanic part, but still relevant to the game at large.

Neuro Twins (as VTubers) fans will love it!

AI content enthusiasts will love it!

Tortoise doesn't need to think as much, to get things done, and even gets forced testing that produces real re$ults, as a byproduct!

All while still being able to say the original game never released, but since it got replaced, it never had to, AND Tutel keeps the title against YanDev, for Indie Vapourware Dev of the Century. So, it's a win-win-win, all around!

gusty bane
#

And then it gets banned on steam for having AI generated content neurowheeze

carmine torrent
#

uh neuroLookUp i havent read allat yet

wanton venture
#

Nah, it's AI-assisted, because he obviously needs to edit it

carmine torrent
#

but that is
crazy

#

actually crazy Tutel

wanton venture
#

In a good way?

carmine torrent
#

uh

#

meh

#

no

wanton venture
#

Why not? What do you think is wrong with it?

carmine torrent
#

like that's cool but not to be expected of Vedal sip idunno...

violet wren
#

classic schizo thread

#

good idea tho

carmine torrent
#

isn't that why he got the GameJam idea?

zinc geode
#

Vedal would prefer working on it on his own i feel I guess he could ask neuro for ideas neuroShrug

carmine torrent
wanton venture
wanton venture
gusty bane
wanton venture
carmine torrent
zinc geode
#

I mean I’d assume they’re not allowed to influence anything directly, it would have to go through vedal

gusty bane
wanton venture
# carmine torrent that seems unrealistic

He's literally demonstrated that Neuro can tell stories.

Edit out the loops, modify it a bit for context, and fix inconsistencies, and he's good.

ChatGPT can write stories, too. I know people who legit made their resumes in ChatGPT...

And that provides an excellent testing and development sandbox for the Twins, too

wanton venture
carmine torrent
gusty bane
#

https://youtu.be/tYxZ_LOlFnU?si=cM1xHmkV15LAhemk
Heres an example. Theres also Ludum-Dare

YouTube
Game Maker's Toolkit
The Best Games from GMTK Game Jam 2023

Full GMTK Game Jam Playlist - https://www.youtube.com/playlist?list=PLc38fcMFcV_vnAZjugCRdKr8_8d_y_rRl

The GMTK Game Jam for 2023 was - once again! - our biggest yet. In this video, Mark runs down his 20 favourite games, from a jam where every game fit the theme “Roles Reversed".

Support Game Maker's Toolkit on Patreon - https://www.patreon.c...

▶ Play video
wanton venture
#

Even if that's true, doesn't mean he can't do the same thing for AA

carmine torrent
#

no the GameJam is for the community

gusty bane
#

So from the little Vedal has said about the game-jam:

Neuro says a game jam topic.
We make games around that topic within a certain amount of time.
AI tools can be used.

wanton venture
#

Oh, that's interesting.

gusty bane
#

If someone could find that clip from a long time ago where he said these things that will be great, its from a dev stream I think like a couple months ago.

violet wren
#

if he's gonna develop game all of a sudden he's gonna have no time for developing neuro. let him take his timeYES

carmine torrent
#

we all should leave it to Vedal

gusty bane
#

If he wants a fan-game made hes probably better off asking his community to do it, cause if there is an organized effort to develop a Neuro fan-game I'd be willing to help, and I'm sure many more will be too.

wanton venture
#

I always want to see the Twins get dev time, so I thought it was a neat way to integrate their development with something else I know he supposedly wanted to do.

unique crane
#

this doesnt seem a good idea

versed hinge
#

Neat idea, but why? Although Neuro can tell stories, they are usually very basic or nonsensical. Also, Vedal would have make new enemies and assets to match the story. Plus, he would have to create cutscenes or something similar to communicate the story, which would take away time from Neuro development that he is focused on. Just doesn't seem worth it tbh.

wanton venture
# versed hinge Neat idea, but why? Although Neuro can tell stories, they are usually very basic...

Neuro can tell stories, but they are usually very basic or nonsensical
Precisely! Doing this would be a way to test and improve the Twins' storytelling abilities and work on their memory (while also progressing the game)

Create cutscenes or similar to communicate the story

  • On the one hand, this isn't necessary, from what I'd seen. The game was pretty basic, in terms of interface. I've only watched what was in the streams, but it could easily be handled by dialogue bubbles
  • On the other hand, this could be a perfect opportunity to add additional skills to the Twins, such as at least basic drawing capabilities. There are already a couple other AITubers out there who draw. This gives the Neuro Twins another competitive edge (and helps move the game along)

It's worth it, because I am suggesting he NOT isolate their development to work on the game, but use the game as a way to further both interact with them and also train and improve their current and even new skill sets.

Idk, maybe my intention wasn't obvious, from the get-go. I'm saying:

  1. Vedal is working on several projects, 1 of which is this game that looks to be going nowhere; and another of which is the Twins, who people care about and want to see improve.
  2. So far, working on either detracted from dev time working on the other(s).
  3. Online culture develops fast, and the Neuro Twins are now no longer the only AI streamers (though, still the most popular, by a tremendous margin), but the skills presented by some of the other AI streamers are different than what the Neuro Twins offer.
  4. By utilising development of current and new skills for the Twins, Ved may also be able to progress with his game, without actually detracting from the Twins, at all.
  • If he has a clear goal of what he is trying to do, development will be easier. His goal becomes Have the Twins co-create a unique and novel game, that not only acts as an additional source of income, but is also sentimentally integrated with the greater and more popular community of my work, thus making it essentially Neuro-sama merch, which more people are likely to buy, with the huge twist of fully character AI co-developed media, and (most importantly) it improves the Neuro Twins and even adds further abilities to their portfolio/repertoire.
  1. Improves his own skills, and acts as a testing ground for ideas he might have had for the Twins, but shelved, due to not knowing how to use them.
  2. Improves the Twins' skills.
  3. Makes the Twins even more amazing, and adds sentiment to the game.
  4. Adds another Lambo to his garage.
#

Nothing about this idea should detract from the Twins' development. It gets them more experience talking, having coherent conversations, making up novel content, and possibly even learning new things like drawing.

And any development to get them to be able to do this is development spent on the Twins.

If he spends 40 hours on the Twins, but he has no clear goal of what they should learn, it might only be like 15 productive hours.

If he uses his game as a training prompt, if you will, he knows what he's trying to do with their skills, and 40 hours may be split into 30 hours giving the Twins new skills or improving existing ones, as well as dev time for experimental stuff that could help them, and the 10 hours spent on adding the content to the game becomes merchandising for them, even if somewhat indirectly.

So, instead of 15 hours productive time and 25 hours of trying to figure things out (net 15 hours on the Twins), it's 30 hours of production time and 10 hours of indirect Twins merchandising (net 35 ± 5 hours on the Twins, depending how you calculate "merch" to weigh in).

This is part of a key AI development strategy, btw: Task oriented training

versed hinge
#

He seems pretty productive to me.

wanton venture
#

(And let's be real, who doesn't wanna see the Twins work together with him on a project? There's so much sentimental value in the idea of them working together on it)

wanton venture
# versed hinge He seems pretty productive to me.

I'm not disagreeing with that. He's doing quite good with the Twins, but I'm trying to say that he can do better, by leveraging his side project as part of their development.

Could be after the holidays, maybe not even until next Summer. I'm just saying, it's a way to move forward on multiple fronts and really bring the Twins to a completely new frontier.

versed hinge
wanton venture
#

Also, why do you say they'd need GPT to do better? Their language models aren't bad at all.

And if he uses some of the stuff I've linked him (assuming he is not, already, but as we all know: DMs Allegations™️), I included several Arxiv things on improving memory, developing context, reducing GPU load for training, etc.

As well as neuroscience stuff related to cognitive psychology, which can be used as further integration for their development.

He has plenty of resources, if he looks at them, that could get the Twins to GPT-4 level or higher, without ever touching prop code from those sites.

#

This is an opportunity to make the Twins incredibly advanced, with task-oriented training, by leveraging his side project as the prompt (both for himself, to focus development; as well as for the Twins, to focus training).

unique crane
#

🤨

faint whale
#

The problem with the idea is that is too much work,if there was a Group of People to help it could be tried but alone nah

versed hinge
#

Vedal is a one guy, not a company full of educated AI programmers

wanton venture
#

Is it not obvious that all this requires is for him to continue working on the Twins, but use the game to focus what he builds into them and how he trains them?

Ok, how about if I phrase it like this...

Should the Twins be more coherent in conversation?
Yes.
Does this help with that goal?
Yes.

Should the Twins learn to tell stories even better?
Yes.
Does this help with that goal?
Yes.

Should the Twins learn to draw?
Possibly, it'd be both cool and competitive.
Does this help with that goal?
Yes.

Should the Twins' memory be improved?
Yes.
Does this help with that goal?
Yes.

Should Ved experiment with new training and model engine designs, to enhance the Twins in other ways?
Very much most likely.
Does this help with that goal?
Absolutely.

Does Ved sometimes need re-focusing and runs out of ideas?
It seems so.
Does this help him direct his attention?
Yes.

Would it be cute as hell, if the Twins helped their dad make a game?
So. Fucking. Adorable. Hell. Yes.

#

This is perhaps one of the few ways to bring the Twins to a new level, without doing anything differently, except that he has an idea of how to direct the process -- it's not unclear, often random testing cycles

unique crane
#

they are still AIs

#

making a game would be mostly vedal, adding ability to code to neuro isnt as easy as you think

#

the best they could do is give ideas, and most of them "wouldnt make sense"

#

adding this and that isnt really easy
the ability to draw, better memory, etc isnt something that can be done in a month

#

this would include neuro knowing artistic styles, remember them, trying to legitimately draw something (reminding you it hasnt been achieved without training data from real art) and know what she drew

#

telling stories is a matter of finetuning the language model, and provide a better "input" to the ai (ex: tell me a story about x)

#

trying to get an AI to draw without real data from artists is something we havent been able to do, and vedal said he is trying not to use data from other artists, making it currently impossible for neuro to draw

#

They are not humans, they will never be, the closest to a human we will get comes from data from other people, Neuro isnt able to create new things from data she hasnt gotten.

#

I hope you understand Neuro is more of a tool than an actual Chatbot

#

atleast currently, it will hopefully change in the future

versed hinge
hoary lichen
unique crane
#

the way to beat a schizo is to be more schizo neuroTroll

hoary lichen
versed hinge
#

Another essay incoming

wanton venture
# unique crane They are not humans, they will never be, the closest to a human we will get come...

There is so much to unpack here... I feel that my messages, despite all the effort put into them, are getting like... 1 out of 6 words actually read...

Ok, lemme try to go point by point:

Making the game would be mostly Vedal
I did not actually say it wouldn't. I said he could use them to assist him in mostly the story, and this acts as a way to figure out how to improve their models, train the models better, and (as you put it) fine-tune their models.

Adding the ability to code to Neuro
I never suggested that, at all. If he did that, it'd be fucking siccc, but I never suggested that. If he goes that route, it's 100% on him, and I'm not pushing for him to do so.

They could give ideas... Most wouldn't make sense... Fine-tuning their language models...
Yes, exactly. It acts as a way to figure out what can and should be fine-tuned; helps him to work on them becoming more coherent; and from the beginning, I said they'd give him ideas for dialogue and story.

Nothing I said contradicts this point. I don't know how I phrased it to indicate the purpose was not in improving their language models (among other things).

Trying to get an AI to draw without real data from artists is something we haven't been able to do
I have no idea where this came from. I said he could (possibly, optionally, but ideally) add a drawing model to them. I never said anything about where he gets the data from. I even indicated the process is ideal and very welcome, but mostly optional.

Further, multi-modal cross-training of her vision model while watching things with partially supervised learning of the association of images with what he calls them and text validation can enable:

  • Makes drawing model
  • Improves speech recognition
  • Creates context-relevant associative weight matrix between audio, vision, and language store models

Neuro isn't able to create new things from data she hasn't gotten
This legitimately goes in line with my theory of how to optimise cross-training mechanisms.

Humans can't really do this, either. We "imagine" things from experience, and the memories get jumbled in our brains, and on retrieval, the data corruption/lossy compression of encoding creates "new" ideas. The brain itself is an associative matrix...

So, yes, she can, if the associative matrix is properly developed to enable cross-trained multi-modal systems to interact in her AI. I'm pretty sure he is doing something similar to this, with the new interactive Minecraft model, tbh, where he said Neuro can take or decline suggestions on what to do.

unique crane
#

i feel like i shouldnt have talked bere

#

Edelweiss, you explain too much and eventually get into unnecessary stuff. Try telling me this in short messsges:
What is the game about?
How would Neuro be able to help?

wanton venture
# unique crane atleast currently, it will hopefully change in the future

My theory goes like this:
Using multi-modal I/O, train the AI to use external tools (including APIs and sensors), and build them to fill in a series of association and shortcut models.

Her LLM is a tool, yes. Her drawing would be a tool, yes. Her speech would be a tool, yes. Her game AI would be tools, yes.

And the AI system connecting them, and trained to use and respond to them as I/O, are the full entities, that are Neuro and Evil.

This goes way beyond the scope of the suggestion, though, and is far more architecture of how he could improve them in a massive reconstruction.

I am trying to focus on baby steps, and the game seems like a great way to work on it.

wanton venture
#

The game itself is an aside
(READ TITLE: ABANDON THE ARCHIVE)

unique crane
#

Okay, again in a short message, how would Neuro evolve by doing this game?

wanton venture
#

Also, I forget who said it can't be done in a month, but I don't wanna scroll that far back up. I never said it could. The fact the release is supposedly 3 weeks away, and that he posted that, just gave me the idea.

#

Steam egging Ved on to finish his vapourware

But what if...

What if he used the game (and prepaid Steam store page slot, BC you pay before listing it...) as a way to work on Neuro, but with a goal in mind?! vedalHUH

wanton venture
unique crane
#

There we go, you managed to explain your goal in an understandable way that everyone can understand and not think of you as a schizo person

#

I would say thats a cool idea, it would help vedal tune Neuro for storytelling, allowing her to sound more human

wanton venture
subtle adder
#

Personally, this seems to be a classic case of "Awesome in theory, kind of a mess in practice" to me.
Sure, it's most definitely a neat idea on paper - and something I'd love to watch actually happen - but I'm not gonna be holding my baited breath for Vedal to execute on this...

Time will tell.

grizzled trail
#

This ^
Im pretty sure Vedal is busy with doing actual dev work on other things he wants to improve. Plus, he has a list with all the things he wants to do. Even if he agrees with this and adds it to the list, I dont think this would be so high in priority because the twins already have enough capabilities for streams to tell a story (as demonstrated by the witch/Halloween stream), making work on better storytelling not that efficient comparatively. If AA is actually done gameplay loop-wise and only needs lore for the game then I could see this happening, but right now its unlikely because of the upcoming subathon and Vedal working on other, more Twin-centric things.

wanton venture
# wanton venture Ok, I think I condensed it: ***The goal is NOT to finish the game. The goal IS ...

How are people still misunderstanding what I was suggesting..?

I even managed to finally trim down the explanation, (here)

AA is not the actual goal, but a mechanism for further improvement of the Twins. Anyone who says the core and primary purpose of this thread is to finish AA doesn't get what I've been saying this whole time.

I watched the improvements they made in the Halloween Stream, as you said @grizzled trail , and that was actually part of the inspiration for this idea.

Neuro is apparently pretty good, so far, when it comes to coherent very short narratives.

This was supposed to further improve on that, and help Vedal use a task-oriented approach, with task-driven training, to develop the mechanisms for more complex long memory (including improved inter-session context), as well as things like help the Twins to learn to divert and reconverge focus between multiple tasks.

The idea was wide-ranging and has very little to do with AA, and absolutely no timeframe.

He could literally use the game as a year or more worth of increasingly complex training tasks and challenges to himself to improve their models and the engine code running them (of course, I don't know how much of their code is stock from Torch, and how much is drastically modified, but the point remains...)

The goal is to use the game as a set of objectives by which to focus training and development.

Finishing the game is entirely optional and nothing more than a nice-to-have.

The expected time limit is anywhere from as soon as I post this, until [Heat Death of the Universe | it gets released and he's sick of updating it].

I am saying, Don't just teach her the A, B, C's by having her write them down, over and over; have her learn to sing them to Twinkle Twinkle, like we all did, because it's easier to remember

Co-creating the game with them is EXPRESSLY for the purpose of guiding development (where the game may facilitate doing so), while providing a more efficient training environment for them to develop deep context awareness. If he actually completes the game, that almost doesn't matter at all.

#

You know, actually, re-reading what I said...

While the process can still be used as training data, I actually am not sure if they have any other learning algorithms outside of GPT-like...

I kinda forgot that GPT-like is really what most people use exclusively, for this sorta thing...

If they only can be trained in an offline state, while it may still be useful, it will be nowhere near as effective as what I was talking about...

I might be confusing their design too much with my own thesis...

grizzled trail
# wanton venture How are people still misunderstanding what I was suggesting..? I even managed t...

Im not saying that you're trying to finish AA, im saying that your view literally means ditching the game to make the Twins better and maybe use AA as a display of the twin's progress.
That's objectively bad because:

  1. The actual AA only fans will be disappointed (case point: The shared mural in the forum, the AA fans literally have their own corner begging Vedal to actually work on the game).
  2. (Im repeating this again) improving a already working feature isn't that efficient or recognisable as doing other work like fixing the Twins responding to spam way too often.
    Yes, your idea shouldn't be high priority and this could eventually be worked on, but i dont see it done in the foreseeable future since the tutel has STILL many things to work on, like the Minecraft AI, the Among Us AI, Slay the Spire AI (even tho its debatable whether it would be higher priority than this).
    The point you're trying to make is that the Twins should have an upgrade to make longer stories, right? How would that be implemented as part of a stream? An entire 2h stream dedicated to only like 5-6 stories? I doubt that would hold any viewer attention and Vedal wouldnt use it again.
    This entire last paragraph is what i exactly mean by "inefficient" btw
wanton venture
#

I'm sorry, I'm really sorry. I'm like cackling at myself for how much of a projection this whole thread was, now.

I spent so much time researching and working out the abstracts of my own thesis, I completely forgot that I made this design in the first place, and it's not how current AI systems even work. I had myself believing they were a trimmed down version of systems compatible with an entirely different learning design I spent several months fleshing out.

Even if Tutel read everything I ever sent him, his use of it is almost certainly different than what would benefit most from this.

Schizo threads for weeks
I might be, at this point. I forgot my thesis is for an entirely new and unrelated ML paradigm.

@grizzled trail I'm not sure, anymore, actually. Now that I realised the above, I have no idea how much they would be able to build the interlinking knowledge basis that Longer stories <=> Role playing <=> Multiple conversational contexts.

I have no idea how he made them, but I forgot it's almost certainly nothing like my architecture evilWiFi , so maybe it really was a bad idea.

I should post fewer suggestions, related to their development, I think. Tutel won't do any of this, and he literally can't possibly know what I'm talking about, BC it's not even a published architecture, yet.