#Project: SURPASS Plan 0

1 messages · Page 1 of 1 (latest)

hollow sigil
#

Project: SURPASS

3D EXPOSURE THERAPY GAME

DOOM. Doom is designed with (mostly) 3D world spaces and 2D NPC art.

Neuro and Evil are already essentially ready to be dropped into original Doom to experience easing into 3D spaces from their mostly dark and flat computer holes.

Rather than building unique code to approach the problem of the inevitable challenges that will meet the girls on transitioning from 2D to 3D, why not just let your AI do AI!
AI logic says we always have room for another black techno sorcery box.

So just drop them in and let them stream DOOM, 2D in a somewhat convincingly 3D land of demons and murder.

They'll be fine

#OGdoom
#transitioning
#"Project: neuroHypers Neuro SURPASS"neuroD

languid cypress
#

first of all I think this would've been better if you put this in the original thread so you dont clog up the channel and so that its easier to follow on from what you said

second it's not really a "neuro integration"

third since there's a lot of real-time-ness in DOOM, you kind of need a second AI to interface with it anyways, at most you can use the API to let Neuro direct the AI on where to go and what to target but at the end of the day I don't think DOOM is a good game to try and integrate

mighty fjord
#

depend what you plan
from one side is what is above said. Neuro/evil as commander and you need an ai between to take care of real-time elements
but from other side, the first DOOM is one of the most modded game that exist. Its much better strategy to look for things that exist and work on this. There are definitly tons of posiibilities what would made working for in solo-play or coop with vedal

#

shortly:
basic game - no
modded - there is possibility but need to find ones that work

hollow sigil
hollow sigil
# languid cypress first of all I think this would've been better if you put this in the original t...

I can appreciate constructive criticism, so thanks. I especially appreciate just the time and thought in general you've shared here.

Perhaps I did use the wrong tag for the two different projects, and I definitely preferred having them both nested within one overarching category header, but Discord just doesn't seem to want to do that in an appealing manner for me. I'm sure I'll probably make the thread heading a description of the broader terms and purpose, a mission statement, as the beginning of the thread next time, and individual experiments should snuggle together nicely below.

If you really think people will have trouble engaging with the channel because of my amateur application due to lack of experience with Discord, maybe a mod would like to fix it, because I think I did ok. I have a lot of different things on my plate.

Once again, thanks for engaging.

Sincerely, Jim

hollow sigil
gleaming echo
hollow sigil
# gleaming echo The issue with 3d games (or more complex 2d games) is that their vision is bad s...

There are a couple idea behind this: Once is providing transitions from 2D to 3D spaces through games that leverage both. The thought is further that if we're to treat these models as conscious, regardless of whether they are, then we should say least try some more human-like training alongside the typical data-slurping they're getting.

It seemed to me TES 2 might be a good possible game to work with because it has similar 2D/3D usage with mostly simple 3D landscapes traversed by 2D sprites like what the girls already are. And it's quite modable with Unity especially.

Did that make sense?

hollow sigil
#

@gleaming echo

gleaming echo
# hollow sigil There are a couple idea behind this: Once is providing transitions from 2D to 3D...

Well you still need to take the game and turn it to text (as an LLM can only do that) which for 3d enviroments is really hard and you need to make it so that they can interact with the game in text which again is quite difficult for 3d games. I assume minecraft, which is only 3d game they have, uses baritone which completly takes care of pathfinding and movement they give it coords and it moves for them however thats possible cause MC is simple and on a grid its also somewhat simple to automaticly turn it to text like there is chest at x y z or a player and interaction is the same like hit player and baritone does all the moving and hitting for them.

hollow sigil
# gleaming echo Well you still need to take the game and turn it to text (as an LLM can only do ...

Didn't the girls have access to others' APIs? I was thinking the game could be running on its own server, just dealing out data as needed. I am still beginning to visualize it.

Anyway it seems like you're imagining that they would have to process live or convert fully the game ahead of time, but I think it could be done on the fly from another server.

Another problem is that we don't really know what is going to be available from the black box side that Vedal runs. There would be tools there we probably won't get access to, so maybe you're right.

gleaming echo
hollow sigil
#

I'll be considering this in depth. Much as my brain can

#

I guess that, just because this conversation feels incomplete without mentioning this thing I forgot, both the games I was thinking of are just barely 3D, and not all the time.

Technically I don't think they even have any 3D information until you go indoors and have some ceiling or walls around.

The point being that it's the least technically demanding @gleaming echo ... I think this covers at leave a portion of your memory concerns, and maybe will make in it easier to break any potential tasks into smaller, more manageable bits.

Did I miss anything else of your responses?

Thank you SO MUCH! BTW 😅

gleaming echo
hollow sigil
# gleaming echo TES 2 or doom while being kinda 2.5d rendered still are still a 3D world and th...

That I understand - sorry for not better addressing it.

My thought is that modern games graphics engines generally (in part) simplify by only showing the imagery viewable from the gamer's POV. If an image were generated every in-game, say meter, it could tie the visual to the x/y/z information alongside it.

I'm not even in the intellectual ballpark of somebody like a developmental neurologist or a computer scientist specializing in AI learning, the type who might have a very specific idea of what we should be testing and probing into the brains of an LLM in the pursuit of teaching it 3D pathfinding similar to humans', but you've given me a lot of useful insight that I'm going to think about.

Can't thank you though for your time NeuroCorpa

hollow sigil
# hollow sigil That I understand - sorry for not better addressing it. My thought is that mode...

I just wanted to share a bit more of my thinking so I'm hopefully better understood.

Simply, if we're teaching Neuro-Sama, then we teach. It feels like nobody is planning to try AI training organically, but instead just assume real artificial intelligence won't need the same time and nurturing as anyone else.

Shortcuts are fine just like with normal human children, if they lead to the right goal, but it's not simple absorbsion that makes a person a person.

So this project idea is meant to take the idea of this LLM being more akin to an AI, and on its path to full personhood.

That means we should give her time and practice like a human child.

I suggested OG DOOM & TES2 because they are really just 2.5D, so good for practicing transitions between different environments.

And they are both primed for modding and have relatively simple structures for integration.

I'm not sure why text would have to be continually pumped in on the fly, but if that's true, it's easy to imagine, TES2 especially, feeding in the info at a hyper-zorkian pace no problem.

thin hamletBOT
#

aww, thank you~ ( ◡‿◡ *)