c0de517e/AngeloPesce

Why nothing great is ever good c0de517e.blogspot.com/2017/10/the-cu… - Over-engineering is the root of all evil c0de517e.blogspot.com/2016/10/over-e…

Does anyone here know anyone who knows people at Pearl Abyss? DM me. Organizing REAC 2026...

@gabriellkann @SebAaltonen Now of course it's not that cut and dry and Sebastian is totally right that there is still a ton of space for engines to exist, maybe not-quite-as-generic but not 1:1 to games either, and so on. Tweets are not the space for nuance

@gabriellkann @SebAaltonen To be clear, a game needs an engine - so ofc when I say that you can make your game, I mean both. But it seems that if we believe AI speeds up engine dev a ton, then it also makes sense that engines per se (i.e. not 1:1 to games - but generic) make less sense. Right?

Now is the perfect time to write your own engine. The argument in the past was that you don't have time to write all the less important generic features. Now, the LLM will do that for you. You can focus on things that differentiate your product.

I'm finally writing up how Nanite Tessellation works. The first few blogs posts are up. More will be coming. graphicrants.blogspot.com/2026/02/nanite…

@SebAaltonen Meh - it's not a product, it's R&D stuff - still a long way to be something useable, but at the same time, can't be ignored.

Why do all Genie 3 videos look like ultra slow motion? Like the system is lagging badly, character moving underwater, etc. Is this because it's 24 fps and faster movement would make it stutter badly?

World models hallucinations. c0de517e.com/027_hallucinat…

Amiga OpenGL remote rendering. OpenGL commands get serialised on Amiga, sent over a TCP socket to a remote rendering server. That is Amiga 68k GLQuake (running in winuae) rendering on my PC RTX3080 GPU :)

Here’s an interesting new JCGT paper showing that small neural nets can outperform LUTs in speed/memory/accuracy even when embedded in purely CPU-based renderers. The authors are some ex-Pixar rendering guys who are now at Aurora. jcgt.org/published/0015…

My #SIGGRAPH2025 lecture recording is now available for all. Enjoy! youtu.be/VTrdeqMMMK0

@Luckyballa The main downside (in general) is that differentiability is an extra constrain on the format, that helps only the encoding step, not the entropy.

Encoding speed is the main downside of differentiable representations, and that’s what keeps them from being useful at scale I think it’s going to change

@KatarinaNoKami @just4leaks2 It's much much cheaper, but it works only for planar reflections.

@just4leaks2 Isn't this more or just as costly as Raytracing, if it were done with new games ? just asking btw

Reflections without Raytracing Ever wonder how they had perfect reflections without RTX? They just built the room twice and flipped it. 🤯 CoD Ghosts: NextGen Testmap

Meanwhile Suslik (Alexander Sannikov) is improving Radiance Cascades for the general 3D case: youtube.com/watch?v=rG2aok…

@iquilezles @ID_AA_Carmack No paper needs to be a paper. But if you are an academic, you have to publish papers because the system works that way and we have not found a better one to replace it yet.

@ID_AA_Carmack I mean - MANY computer graphic papers didn't need be a paper to begin with, a quick tweet/blog post/github project/shadertoy demo would have sufficed. Imagine if all the time our collective brainpower wastes playing the academia game would gone into investigating the next idea.

Would papers get rejected if they used legal code with variable names instead of Greek laden pseudo code?

@ID_AA_Carmack Lots of hate in the thread. The answer is no (usually), pseudo-code is not mandated. But length/width is a real constraint. Also, everyone has its language. Of course professional programmers are more comfortable with code, while some academics like pseudo.

@tomwarren It was absolutely obvious from the get go, but sadly Twitter now is all about attention farming and the algorithm promotes only noise and blue-checkmarks

Microsoft isn't rewriting Windows in Rust with AI, despite many reports about it over the past couple of days

@rohanpaul_ai "my goal" != "microsoft's"

wow. Microsoft wants to use AI to wipe out all C and C++ code by 2030 and replace everything with Rust. Their new “North Star” metric: 1 engineer, 1 month, 1 million lines of code. Microsoft Distinguished Engineer Galen Hunt writes in a post on LinkedIn. “My goal is to eliminate every line of C and C++ from Microsoft by 2030. Our strategy is to combine AI and Algorithms to rewrite Microsoft’s largest codebases. Our North Star is ‘1 engineer, 1 month, 1 million lines of code.’ To accomplish this previously unimaginable task, we’ve built a powerful code processing infrastructure. Our algorithmic infrastructure creates a scalable graph over source code at scale. Our AI processing infrastructure then enables us to apply AI agents, guided by algorithms, to make code modifications at scale. The core of this infrastructure is already operating at scale on problems such as code understanding.” --- Source Linkedin post linkedin .com/posts/galenh_principal-software-engineer-coreai-microsoft-activity-7407863239289729024-WTzf/

@miketuritzin advances.realtimerendering.com ?

It's hard to find videos of SIGGRAPH "Advances in Real Time Rendering" talks, but many of them are freely available: Go here, and click into "Courses" for the year you want, then one of the "Advances ..." sessions. dl.acm.org/conference/sig…

@joseph_h_garvin Agreed, for security-critical code the equation is different.

@kenpex It's much greater than zero because 1) those remaining bugs take disproportionately long, like you said 2) those remaining bugs create security vulnerabilities 3) memory safety implies data race safety which is even trickier We have data too, most CVEs are memory safety issues.

This is why I was never excited by static memory safety guarantees. My code does have bugs - plenty - luckily, I catch them while testing - good. The % of my bugs that are memory, or even threading related is vanishingly small. Can remember only one in the last year...

@olson_dan All true, but nothing that invalidates what I said?

@kenpex I highly value static memory safety checking. Memory and threading issues: - hide until data changes trigger them - exhibit non deterministic behavior once discovered -require specialized debugging techniques to find and fix