it's hard to determine the Orb Levels with all this fog

we probably shouldn't be using the tool made by the people who are afraid of Roko's Basilisk

reading docs for SIMD intrinsics and wondering how we ended up here

it's hard to determine the Orb Levels with all this fog

everything seems fine

everything seems fine

more testing

current status at the office of geometry

wallpaper vibes

wallpaper vibes

changed some shader parameters around [orb levels: nominal]

the intern is setting up a new experiment

this turned out slightly better

some more glass panels

it's not hard to write a raytracer. it's hard to write one you want to use for more than a few minutes

this turned out slightly better

indirect fog light #raytracer test

there appears to be a strange divide between: genAI / no (gen)AI / genAI, but only for coding for some reason. and i feel a little bit insane

aaaand one more

more testing

more testing

volume light on/off

the obligatory teapot test rendering. now with better adaptive volume sampling

i really want elon to find out soon

fixed some more issues with the adaptive volume sampling. it's hard to see any major difference between this and similar older renders. except this was done in less than half the time. with more adaptive sampling. (unless i've made mistakes)

i have made myself a really strange bug that may or may not be a bug

the glass has nice distortion. rendered in custom raytracer #graphics

i added a ray counter to my raytracer, so i can tune performance better. the right most columns show rays/pixel, "pixels processed", rays/second, total rays

'for an estimated 4% of people, this mental imagery is weak or absent. When researchers ask them to imagine something familiar, they might have a concept of what it is, and words and associations might come to mind, but they describe their mind’s eye as dark or even blank.'

the glass has nice distortion. rendered in custom raytracer #graphics

this scene again using the most recent adjustments to the monte carlo integration. it rendered in 69 minutes 12 seconds. nice. #raytracing

this scene again using the most recent adjustments to the monte carlo integration. it rendered in 69 minutes 12 seconds. nice. #raytracing

i liked it better when people didn't care that much about computers in general. or at least i preferred it over everyone being enthusiastically wrong about computers everywhere and all the time

volume light on/off

software is going to become so much more frustrating

Physical Chaos 物理的なカオス (Chaos Theory カオス理論) Making chaos with physical simulation 物理シミュレーションでカオスを作る kyndinfo.notion.site/Physical-Cha... #p5js #math #physics #code #art #design

🕯️ 🕯️ 🕯️pop 🕯️ 🕯️ 🕯️

i think i found a way to make the sampling density in my monte carlo volume light integration depend on distance along the light path. so i can sample less for rays that are "far away". to speed things up while maintaining some detail. might have some test renders soon

i put the load and current memory use in my shell prompt. don't know why i never thought of this before

its not new that you should consider whether you trust a service before giving them your data. but doubly so now that almost anyone can make a service-shaped artefact on the web in days. complete with every imaginable security hole and rookie mistake

at least we can agree that more meetings should have been emails

this API requires a GPU ...

things, i gather, have gone out of control

i liked it better when people didn't care that much about computers in general. or at least i preferred it over everyone being enthusiastically wrong about computers everywhere and all the time

i think this agument (dev. speed) can be applied to software dev in general. and possibly also to explain why some are so eager to use coding agents: they have bad tooling

you can stack or layer LLMs any way you like; you are not going to create conciousness. no matter how much you want it to be true

computers talk to each other all the time, and have done for decades. having them use natural language is just a step back in terms of efficiency and security

i like the tiny rainbow in particular

qmv is good for renaming multiple files in the terminal. it opens the files in two columns in an editor, and you can edit the file names in the second column

another rendering of the cube monument for [redacted] cube monument reasons