|
 |
Hi(gh)!
Reading an article in the German new portal "Zeit Online"
(https://www.zeit.de/digital/2024-07/ki-meta-llama-3-1-newsletter-kuenstliche-intelligenz),
I stumbled upon Mark Zuckerberg's new AI system, Poe 3.1: https://poe.com/
After having extensively played around with graphical AI systems such as
Stable Diffusion, this is my first experience with text-generating AI
systems (like the famous-infamous ChatGPT)!
I heard before that such AI systems were pretty capable of generating
code in a vast variety of programming languages... POV-Ray is a
programming language, so I tried the archetypical basic scene with
which, according to conventional wisdom, every budding POVer begins: a
reflective sphere over a checkered plane. For me, this is my official
debut scene delayed by 29 years (when I discovered POV-Ray in the first
place), or at least by almost 24 years, as I posted my first scene here
(anyone remembering "Afghan Sunrise"?) on September 21, 2000... I'll
nickname this first regular debut scene "2000 Slightly Delayed",
alluding to a piece of guitar-driven German electronic music from 1978...
I entered the following prompt:
Generate the POV-Ray code for a reflective sphere hovering over a
checkered plane under a blue sky!
and, in the blink of an eye, Poe 3.1 generated this piece of flawless
POV-Ray code:
// Camera
camera {
location <0, 5, -10>
look_at <0, 0, 0>
}
// Lighting
light_source { <-10, 20, -10> color rgb <1, 1, 1> }
sky_sphere {
pigment { color rgb <0.5, 0.7, 1> }
}
// Plane
plane {
y, 0
pigment {
checker
color rgb <1, 1, 1>
color rgb <0.5, 0.5, 0.5>
scale 2
}
}
// Sphere
sphere {
<0, 1, 0>, 1
finish {
ambient 0.2
diffuse 0.4
specular 0.4
roughness 0.1
reflection 0.5
}
pigment { color rgb <1, 1, 1> }
}
Yes, I'm pretty astounded... but, for an AI using 400 billion parameters
(according to the Zeit Online article), generating error-free POV-Ray
code should be of the simpler tasks...
When taking a closer look upon the code, I noticed that the sphere
(radius 1 unit) is not really hovering above, but in fact touching the
plane...
And of course I rendered the scene... attached here!
The camera looks down on the sphere, so the blue sky is visible only by
reflection from the sphere's surface... this is not fully precisely what
I intended to have it looking like!
But AI outputs are typically randomized within the limits of the given
prompt (at least I experienced this with the graphical systems)... so
I'll give this prompt further tries! Watch out!
See you in Khyberspace!
Yadgar
--
VBI BENE, IBI BACTRIA!
Post a reply to this message
Attachments:
Download 'poeray_rsocp_001.png' (64 KB)
Preview of image 'poeray_rsocp_001.png'
|
|
|
|
Hi(gh)!
On 26.07.24 18:05, Bald Eagle wrote:
> Well that's pretty cool :)
>
> See what else you can get it to do.
>
> Patterns
> Algorithms in SDL
> renders in the complex plane
> Spirals
> Mobius transforms
...of course a full-fledged photorealistic virtual POVghanistan with
topography detailed down to single grains of sand, river network from
the smallest mountain creeks to Helmand and Amu Darya, complete
ecosystems from viruses and bacteria to cedar trees and snow leopards,
all human buildings from black tents and yurts to Timurid mosques and
palaces (as they were before 1978, later on of course also the
futuristic "Afghatopia 2050" mode!), archeological sites, the entire
traffic network (of course no railway lines in pre-1978 mode!) with
gaily painted "lorries" and imported run-down buses, some 15 million
POVghans of all ethnic groups and social milieus (and of course several
ten-thousands of hippies passing through on their way to India!), each
one with her/his own personality and biography... and than exporting the
whole thing to X3D, making it dynamic and real-time interactive,
implementing all Afghan languages... *rant* *rave* *hallucinate*
O.k., seriously: firstly I will continue exploring the generation of the
most basic SDL scripts before gradually increasing the scenes'
complexity... and even though I hope to be able to make use of Poe (or
other, comparable AI systems) when developing POVghanistan/Khyberspace I
would not like to have them doing all or even most of the creative work
for me - at least with dedicated graphical AI systems like Stable
Diffusion, my experience during the last 12 months is somewhat sobering,
as they still produce lots of hallucinatory nonsense, at least when
using the cost-free versions (or my prompts are not sophisticated enough
yet)...
Nevertheless, AI-guided programming is a massive, fantastic field of
learning which might absorb me for many years to come!
See you in Khyberspace!
Yadgar
--
VBI BENE, IBI BACTRIA!
Post a reply to this message
|
|
