Literal centuries—maybe even millennia—of brilliant effort went into making this thing, and I'm totally fine saying "meh" to all of it because of one typo that's entirely my own fault, and a huge backlog of Overwhelmingly Positive games.

IMO the core of game industry unsustainability is simply: More really great games than people's free time can sustain. I believe Arc Raiders is a really great game. But this one tiny thing is all it took to make it a razor's edge less great than all the other really great choices I have.

Not really complaining; there's a TON of great releases right now; I'll happily play those. But observing—as a dev—how one tiny missed keystroke, in conjunction with a one-size-fits-all policy, had made all the difference in which of those new releases get my money and which particular one won't.

I think of WGSL as human-readable portable object code, optimized for very fast conversion to both SPIR-V and Metal IR, and supporting "linking" via simple string concatenation. That's actually a pretty solid combo; just not the one for primary authoring.

Feels like there's some kind of cute algorithm for dithering where you alternate back and forth between picking colors in Oklab, then computing and diffusing errors in linear RGB.

Because aggregating and selling PII for marketing purposes is part of the business model. None of this was ever for our benefit.

Yes, even the local pizza chain. They're not running an ad business, but are selling to a broker, likely without understanding (or caring about) non-pizza implications.

Nothing will ever top EIEIO. www.ibm.com/docs/en/aix/...

That the joke; you wouldn't combine those two approaches. Cut out the middle step and use a NeRF.

The step of solving GI with AI is interesting; that's only barely tractable today and involves lots of complex/fragile optimizations.