Artificial Intelligence and Us: an open debate

[T0ofic]

is glorybound for real after all? Though it was a joke, kind of an ai... thereby to stresstest the support team and the forum spam report function.
I was obviously misled, my bad!

[^Lele^]

^Lele^ is ai a big part of your not work nowadays? given the fact that you are the renowned and infamous problem steerer
what are your thoughts about AI.. and please don't be elusive if even possible. (I'am a real human)..

I am staring, and wondering what will be.

It's hit and miss still, some of the issues are inherent with the current batch of approaches (i.e. the model), others with the stuff they get trained on (the training set).
However, some have found clever ways around some of the limitations, and all of a sudden implausible tools seem to become extra usable/useful (see here, f.e.).

I don't have a crystal ball, nor do i assume what is coming will be identical to what has gone before: after all there is little to differentiate the *concept* of a modern car to that of a 100 years old one (engine, steering, 4 wheels, brakes), but the new ones are a lot fitter for the purpose.
So i stare and wonder.

[piotrus3333]

important next step would be making the training of models easier and more accessible. if I can teach it doing things my way - I will surely keep it busy.

[super gnu]

I am staring, and wondering what will be.

It's hit and miss still, some of the issues are inherent with the current batch of approaches (i.e. the model), others with the stuff they get trained on (the training set).
However, some have found clever ways around some of the limitations, and all of a sudden implausible tools seem to become extra usable/useful (see here, f.e.).

I don't have a crystal ball, nor do i assume what is coming will be identical to what has gone before: after all there is little to differentiate the *concept* of a modern car to that of a 100 years old one (engine, steering, 4 wheels, brakes), but the new ones are a lot fitter for the purpose.
So i stare and wonder.

"cuebric" ahaha

maybe Camer-On would have been more appropriate.

[^Lele^]

important next step would be making the training of models easier and more accessible. if I can teach it doing things my way - I will surely keep it busy.

I too wonder if it'll ever be a thing.
Nuke shows training can be done with little enough data, at least for some of the tasks.
There's surely still work to be done, but the direction it points to is -quite literally- awesome.

[roysgi]

I did a little experiment and asked chat gpt to write me a maxscript that would create stairs. When I run the script it gives me an error. This is what he gave:
-- Define the number of steps and step size
steps = 10
stepSize = 1

-- Create a loop to generate the steps
for i = 1 to steps do
(
-- Calculate the position and size of the current step
pos = [0, (i-1)*stepSize, 0]
size = [2, stepSize, 1]

-- Create a box at the calculated position and size
box = box pos size

-- Add the box to the scene
addNewNode box
)


I don't really understand this, but is the chatgpt in the right direction?

[^Lele^]

yes, up to the last two lines.
You'll need to give it more direction, iteratively, and it'll get you somewhere.
But you need to know the subject well enough to spot the errors in the first place: it's really good at faking it.

[Pixelcon]

I feel like the biggest barrier to getting something useful rather than just interesting ideation is interface.
A basic GUI with a node-based approach to building the image, altering aspects of it, and a basic canvas to direct where alterations occur could really be a big step forwards.

[^Lele^]

Yeah, i too am looking at the general lack of directability as a key issue.
That's surely down to UIs, but it's also got to do with both the models and the training sets.

If the model doesn't segment an image into parts, it's going to be hard to edit the image selectively later (unless one resorted to traditional means).
If the sets that trained the AI never contained painterly images, but just CGI, it'll be impossible to direct the model towards painterly outputs.

Inpainting is a good approach, but still too finnicky (i.e. it'll change my mountaintops along with the sky above them, because i painted over them.).
Segmentation is great, but it's not always possible (think of overlapping elements.).

Extraction/generation of depth information, (partial) occlusion (f.e. via animated sequences) and so on may help with the above, in the long run (think deep images, cryptomatte, etc.).
Who knows what's beyond the current horizon.

[Richard7666]

https://forum.evolvelab.io/t/renderi...evit-model/496

This is really cool and also worrying. Eventually this thing will be able to understand "change the back wall on behind the counter to white herringbone pattern brick"

[Vizioen]

This is the AI: https://www.evolvelab.io/veras

And the results look quite crazy.

[MarkoStojkovic]

Poured concrete, which is what a 3D-printed house is. I think sensationalizing things gets more attention, which is what is happening with AI. I am old enough to yawn at this stuff since there is always something. So I first caught on to the next great thing when I was told that Flash would take over the internet! Now, what? It is the metaverse. What the heck is that?

Totally agree, too many sensations I saw over time about how this or that will solve all humanity's problems D

[francomanko]

Someone said this which i thought was quite apt..."When i was a kid, the future was robots doing the menial work leaving humans to make the art, reality is that the robots are making the art and we are doing the menial work."

One concern I have with A.I, is the ethics, social awareness and motivation of the people and companies behind it all...it seems like its just another gold rush at the moment.

[Vizioen]

...it seems like its just another gold rush at the moment.

As with anything these days, or...ever. If something can be exploited for financial gain, it will be.

[^Lele^]

In this latest CGGarage podcast, you can gleam what Vlado thinks of AI, and in which way it's going to appear within the Chaos Ecosystem.

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