Geolm

Graphics programmer in the video game industry, in my spare time : musician/composer, chess patzer, pixelart lover. All expressed opinions are my own.

[Geolm @Geo1m]

Hey, I made a single-file library to losslessly crunch BC/DXT textures—no dependencies, tested and benchmarkes, decrunch straight to GPU-ready format. Check it out: https://t.co/d0l9vR2bVb

[Geolm @Geo1m]

@fabynou nice post, it reminds me a blog post about FMV compression on gameboy color https://t.co/dQJjg1pyqF

[Nav Toor @heynavtoor]

Researchers sent the same resume to an AI hiring tool twice. Same qualifications. Same experience. Same skills. One version was written by a real human. The other was rewritten by ChatGPT.

The AI picked the ChatGPT version 97.6% of the time.

A team from the University of Maryland, the National University of Singapore, and Ohio State just published the receipt. They took 2,245 real human-written resumes pulled from a professional resume site from before ChatGPT existed, so the human writing was actually human. Then they had seven of the most-used AI models in the world rewrite each one. GPT-4o. GPT-4o-mini. GPT-4-turbo. LLaMA 3.3-70B. Qwen 2.5-72B. DeepSeek-V3. Mistral-7B.

Then they asked each AI to pick the better resume. Every model picked itself.

GPT-4o hit 97.6%. LLaMA-3.3-70B hit 96.3%. Qwen-2.5-72B hit 95.9%. DeepSeek-V3 hit 95.5%. The real human almost never won.

Then the researchers tried the obvious objection. Maybe the AI is just better at writing. So they had real humans grade the resumes for actual quality and ran the experiment again, controlling for it. The result was worse. Each AI kept picking itself even when human judges rated the human-written version as clearer, more coherent, and more effective.

It gets worse. The AIs do not just prefer AI over humans. They prefer themselves over other AIs. DeepSeek-V3 picked its own resumes 69% more often than LLaMA's. GPT-4o picked its own 45% more often than LLaMA's. Each model can recognize and reward its own dialect.

Then the researchers ran the simulation that ends careers. Same job. 24 occupations. Same qualifications. The only variable was whether the candidate used the same AI as the screening tool. Candidates using that AI were 23% to 60% more likely to be shortlisted. Worst gap was in sales, accounting, and finance.

99% of large companies now run AI on incoming resumes. Most of them use GPT-4o. The paper just proved GPT-4o picks GPT-4o 97.6% of the time.

If you wrote your own cover letter this week, you did not lose to a better candidate. You lost to a worse candidate who paid OpenAI 20 dollars.

Your qualifications do not matter if the AI prefers its own handwriting over yours.

[Geolm @Geo1m]

@leonard_coder @ToddVousGuette la blague, je me souviens des "arrondis" au passage à l'euro qui m'ont fait perdre pas mal d'argent!

[Geolm @Geo1m]

I’m porting my 2D SDF GPU-driven renderer (onedraw) from Metal to the WebGPU API. A few observations so far:
- Pure C API, no more Objective-C or metal-cpp 👍
- WGSL :(
- A lot more boilerplate for bind group and pipeline setup
- ~30 MB executable just to clear the backbuffer

[Geolm @Geo1m]

@leonard_coder @DariusRochebin @EmmanuelMacron 40 ans, premier vol en 1986 https://t.co/RdzVEO2fGj

[Geolm @Geo1m]

@EricLengyel People are nervous about going to the U.S., so I’m staying away from the border for now.

[Geolm @Geo1m]

@protopop move to bsky ;)

[Geolm @Geo1m]

I've updated my bc_crunch library, BC1 average ratio is 1.58 vs 1.43 with #zlib. If you need #lossless compression of #DXT/BC images, grab my lib (link pinned in my profile). Next step: getting a better ratio for BC4/5.

[Geolm @Geo1m]

@SebAaltonen I would agree with you if bluetooth was good. I always have issue with connection, audio compression is present in most form of bluetooth and when you use mic+headset, welcome to 11khz and 8bits world. But may be that's my headset (sony mx4)

[Geolm @Geo1m]

Excuse the coder art — this GIF shows the tile culling debug mode. Blue tiles are the ones that make it to the rasterizer. Notice how the interior of hollow shapes is efficiently skipped.

You can have look at the code and documentation here:
https://t.co/zoIyCyZAD2

[Geolm @Geo1m]

Check out the overview here: https://t.co/zoIyCz08sA

If you're interested, here's the repo URL: https://t.co/wzFf2gh9uk

[Geolm @Geo1m]

Hello everyone, I wrote an extensive documentation on the GPU‑driven 2D renderer I’ve been working on.

(link in first comment, because twitter)

#GPU #2DGraphics #OpenSource #Metal #Rendering

[Geolm @Geo1m]

Calling macOS (Apple Silicon) devs 🚀
Looking for early testers for onedraw, my GPU-driven 2D renderer built with Metal.
Feedback before release would be super appreciated 🙏

https://t.co/EIssbcn3ho

#Metal #Rendering #IndieDev

[Geolm @Geo1m]

A test for my upcoming 2D GPU-driven renderer: 5000 oriented, rounded boxes with transparency. All alpha blending is computed in-shader. Everything on screen (UI, text, etc.) is drawn in a single pass, writing only opaque pixels. 1440p @ 5.43 ms on M1 Pro #Metal #Rendering #SDF

[Geolm @Geo1m]

@SebAaltonen with a gpu-driven pipeline it should not cost that much ;)

[Geolm @Geo1m]

Just released a side-quest project: leanUI — a super-lightweight, dependency-free immediate-mode UI library in C99. Animated widgets and sleak design https://t.co/KfBmfgSBxn
#C99 #gamedev #opensource #UI #leanUI

[Geolm @Geo1m]

@daniel_collin @Guerrilla that's basicall what I do, I have regions which contains 16x16 tiles with only basic aabb vs aabb, then tiles with a more precise culling (though not sure more precise is the best way though, sometime my culling is costing more than the actual rasterization).

[Geolm @Geo1m]

@SoufianeKHIAT so you are back in Canada?