Here's grok 4.6's radiance cascade result after a few small cleanup prompts. Lovely lighting!
Caleb Miller
@cmiller__
Graphics programmer and general UI guy. Addicted to JavaScript and caffeine. Pretty good at plugging in USB Type-A cables.
Joined December 2015
- I just tried this radiance cascade test with grok 4.6 high fast, and it: - actually implemented RC with a cascade merge - used far less tokens (20M vs 50M!) - in less than half the time - at 14% of the cost Consider me impressed! They might have something special here.This weekend I tried getting 5.5-xhigh (with opus 4.8 at creative stages) to build a pretty simple radiance cascade renderer. I gave it papers on the technique + two working high quality code repos. I also put a visual review gate in the loop. It failed big time, repeatedly,
- Surprising GPU optimization result today. My engine does a lot of vertex shader rotations with no translation component, easily over a billion per second. To date I've used mat4 for all transforms, as it's the gl-matrix happy path. I just refactored everything to use mat3 for
- The smugness around not reading or reviewing code is especially frustrating when almost every time I use ChatGPT I experience buggy behavior. Both on the web and native app.
- It's nice when big games make raytracing optional, with baked lighting fallbacks (increases install size tho). An optional fast path is cool. But, real-time GI is truly awesome. Even without a day/night cycle, it grounds all dynamic objects into the world so well. I love it.So cool that my PC has to run real time lighting and get -500% performance to support a day/night cycle that doesn't exist in the base game
