Finally set up Claude to run in a super-locked-down container so we can skip the permission fatigue (which inevitably would've caused me to accept bad commands).
aka: I gave my Claude guy their own room. (A basement room, padded cell, locked door, and I feed them through a slit in the door).
goodnight world
bevels + new terrain generation = win
(for transparency; these are the reference renderer shots! realtime will probably not look exactly the same)
and some atmospheric fog
higher render distance
Pair programming with Claude on laying out chunks in a more cache-friendly way in memory & reducing memory overhead/bandwidth of terrain generation. Now boasting fast startup time and very smooth & consistent frame pacing.
Auto exposure, courtesy of Claude running in the background while I do other things - still needs a lot of tuning, but I'll do that later when I can give it more attention.
the new reference renderer is so good
goodnight
Accumulating multiple samples per pixel in the ray traced path now. (Not TAA, no reprojection - just to get cleaner reference stills.)
In short; this is not a skill worth training. Instead, you should do the thing our parents always told us to do (and then promptly forgot about):
Seeing something on the internet is not enough conviction to say or think something. You need to do the work to get at the truth.
2) Even if you've somehow figured out a way to, that knowledge isn't going to be transferable to the next generation. And the next. And the next. Eventually, it will reach the point where you're reading tea leaves, and then you'll be even more destructive.
If you think you can tell AI generated stuff from human stuff:
1) No, you can't. I've seen so many people lash out at actual artists for using AI when they _haven't_. You're hurting the people you're trying to evangelise for. None of us want it. Stop.
As an aside; since switching from rust_gpu to Slang, shader development has been going a ton smoother.
I suppose it really does help to use something that is both reasonably complete and reasonably documented after all.
(I never got to the point of sharing any Rust code because of the toolchain.)
Now rendering albedo, normal and combined AO/emissive maps in the ray tracing path! Still no water pass as that's more complex to pull off.
Next up: going to start setting up a proper tonemapping pipeline and get some antialiasing in there.
Building the ray tracing acceleration structure on the fly from actual voxel data.
One year ago; the original ray-traced lighting solution I was tinkering with for Cavey. Funny that exactly one year later, I'm still playing with rays.
Runs at 90fps on Deck, pretty goooood
youtu.be/PBa2MQ3WkQc?...
2048x512x2048 GPU voxel ray tracing renderer test.
Oh shoot I guess it’s time to follow it up with something more profound…. lets see….
WGSL is terrible and badly equipped for anything GPU driven. But I still haven’t found a shader language I like, so maybe that makes it relatively less bad.
Subscribe for more insightful video essays or whatever.
(The irony of this being an AI post is not lost on me.)
Whatever the answers are… there has to be something more interesting to do with your time than flood Lobsters and HN with the same rehashed bullish/bearish/moralistic blog posts everyone has read two hundred times.
I am pro-human. Can we just go back to sharing what you’re doing?
There are multiple orthogonal questions about multiple orthogonal things. What are the ethics of using / profiting off crawled data sets? Is code generation actually useful for focusing on design decisions, or a distraction from them? What about the interns & juniors? Is human provenance important?
Anyhow, for my first post… Having an opinion about AI is not a replacement for a personality trait.
It’s a complex, nuanced, multifaceted issue with real world weight. I don’t care *what* your opinion is - even in my field, reasonable people don’t agree on fundamentals here - I care about *why*.
