ai is getting so good it can automate doing absolutely nothing

finished 3rd year please don't let me fumble

you try and fail to mimic macos when you had a perfectly good product before. id pay full price for a windows that wasn't actively hostile to the user.

theyre gonna need 10 billion more GPUs to explore the primordial RL gotcha.

okay I think I finally understand shadow mapping thanks to noino🔸's article: noino.substack.com/p/raylib-graph… Next up I'll expand the terrain to many, many times larger than it currently is and then implement CSM.

another slop post with replies acquiescing blindly in a daze, or shilling some product Just to be clear, Karpathy was already behind as a programmer before GPT came out, on account of him being mostly a researcher working in Jupyter notebooks and python

2026 goals: - stop being fat - become better at programming

x.com/i/article/2013…

game development is going slow ive met a beautiful woman

starting 4th year. delaying unemployment. returning with a new C + vulkan renderer ive been working on. cubemap raymarching with depth texture parallax correction for realtime reflections, in place of screen-space. reflects things off screen and takes <0.5ms. pretty cool!

noisy and expensive screen-space radiance cascades using DTPC probe for raymarching. samples scene radiance in all directions, even off-screen. might be a good inbetween for SSGI and path tracing? also snuck in cascading shadow maps c and vulkan custom gamer engine

some perf work on screen/sphere space radiance cascades, from 5ms to <1ms. ray sampling has a tiny bit of noise so its hard to consolidate cleanly, but i havent got into cascade gather weighting either. total frametime around 4ms. time to make the car game????

Introducing FragCoord: My ultimate shader editing tool!

been rattling my brain for like a week and it seems i have misunderstood radiance cascades. my last post was some secret third thing. here's an attempt at SPWI RC using the same raymarcher as before, with lazy bilateral upsampling (needs work) for merge. no noise or TAA, <1ms

minmax upsampling to try fix depth discontinuity artifacts in SPWI RC, and a little volumetric fog because i needed a break. uni about to start again but i hope i have a little time to keep going :)

i loathe ai slop as much as the next guy but there is a delusional psychosis about DLSS5 on the timeline. 'it looks like shit' it really doesnt. the results are absurd for realtime and if nvidia had dropped this before normies knew what ML was they could buy the planet.

greatly improved stability in motion, compared to last time. i went on holiday and read a lot more GI theory + got the OK to write my honours thesis on it. single shot global illumination using panorama space raymarching and radiance cascades, custom C and gaming vulkan engine.

maybe a novel global illumination solution is what it takes to find employment???? physics lights in a big box. this technique is interesting because the main radiance probe can be as small as 64x64 with minimal quality loss. using a mip chain should give a big performance boost.

bistro slide show. too many triangles. showing off global illumination using colourful glowing cubes again, but ive made some improvements. clear instability over long distances, for which i might fallback to screen space. what a joy to create this for my uni dissertation. vk + c