@demofox Finally got to try it and it's pretty good! We switched to it. Also thank you for automatically atlasing png's and optimizing this generator. STBN was taking quite a while to generate vec2/3 textures.

@zeux @castano Matthew just posted the full piece on Roblox and he has a good explanation why losses are real. Interesting perspective how thin are the margins compared to more traditional FAANG: https://www.matthewball.co/all/roblox2024

@msinilo @anji @sbz Most things in the US are decentralized, handled locally and their quality may vary widely. Around me most of the roads are German-like, but in a few places it's indeed a disaster :).

@BartWronski Some time ago WSJ reported 20$/user loss: https://archive.ph/3WDSv so it would be interesting to see how Copilot margins look like nowadays.

@froyok It's not the highest quality, as it was targeting real-time per frame encoding on prev gen consoles, but may still be better than RGBM like encoding: https://github.com/knarkowicz/GPURealTimeBC6H

@aras @torust Storage is a bit higher, but indeed the results are impressive.

@dominikg @msinilo @unjello @BartWronski @demofox AFAIK lowest one is flat 5% using IP Box, but even 12% or 19% is ridiculous given that it's a tax targeting mostly best payed employees like programmers.

@castano @mrf Yeah, and it's really worth it. Decades ago when I worked on baked lighting artists were tesselating everything like crazy in order to avoid those vertex interpolation artifacts, and least square optimization heavily reduced polycount in our scenes :).

@msinilo @BartWronski Twitter is (was?) one the most important social media in the world, so even without amazing financial results it was still worth a lot. Here's a valuation from Professor of Finance specializing in valuation and he arrived at a similar price as the stock market, so at least it wasn't a completely crazy number: https://aswathdamodaran.blogspot.com/2022/04/elons-twitter-play-valuation-and.html?m=1

@BartWronski @msinilo Fair take. Just one thing if you're not familiar with DCF, (or similar tools) it gives you a nice framework to convert your story about the future of the company (Tesla will take over the car market) into stock price ranges with probabilities. So it's not as simple as the current price divided by trailing earnings and captures more nuances, but at the same time also has more uncertainty baked in (will Tesla take 50% or 90%, will market be 3T or 5T?...).