thinkingQueen

You shouldn’t fill the reservoir with the coffee pot, unless you’re really washing the pot super clean after each use. Better get a proper jug for filling the reservoir, so you’re not putting coffee residue and oils back into the clean-water system.

So you think a weighted die is more random than a fair die? A uniform distribution means each outcome has equal probability; it doesn’t mean the outcome is predictable.

Valve doesn’t release that many games though.

Isn’t PS5 Pro a higher efficiency version?

That’s a fairly unreasonable take. We are talking about a hobbyist Github account reverse engineering decades old hardware. The warning is in italics clearly after the introduction. This is stuff for the experts. What more do you want?

Those are good venues to sneak JPEG XL into the mainstream. It would be a pity if it became another JPEG 2000. On the other hand, JPEG 2000 was probably just too advanced and computationally complex to be widely adopted at the time. Sometimes I look back at it and, after all the extensions and revisions, it feels like it has everything—but it’s still a niche codec. A cautionary tale, and a pattern that tends to repeat itself with codecs...

AV1 is being actively claim-charted by a lot of companies right now, and lawsuits are almost certainly coming. The same process is already starting for AV2, but most players are waiting for the AV1 cases to mature first.

People keep calling the AV-family codecs “royalty free,” but in practice they increasingly look like a legal and financial gamble.

That is a very sour take on Iceye. Based on what I’ve read Iceye is doing better business than Umbra. They can’t be that bad. Also, Iceye has demonstrated very good military performance in helping Ukraine.

It looks like he’s confusing Paul Thomas Anderson with Paul W. S. Anderson. The latter directed Resident Evil, which the author refers to.

Beta 5.2 was when I had the best time with Counter-Strike. de_dust with a Colt was fun. Never forget the AWP snipers lurking near the big front door in cs_assault. There were some weird maps like cs_siege — I think it had some sort of a moving vehicle there somewhere in a tunnel.

You’re comparing apples to oranges.

Daala was never meant to be widely adopted in its original form — its complexity alone made that unlikely. There’s a reason why all widely deployed codecs end up using similar coding tools and partitioning schemes: they’re proven, practical, and compatible with real-world hardware.

As for H.265, it’s the result of countless engineering trade-offs. I’m sure if you cherry-picked all the most experimental ideas proposed during its development, you could create a codec that far outperforms H.265 on paper. But that kind of design would never be viable in a real-world product — it wouldn’t meet the constraints of hardware, licensing, or industry adoption.

Now the following is a more general comment, not directed at you.

There’s often a dismissive attitude toward the work done in the H.26x space. You can sometimes see this even in technical meetings when someone proposes a novel but impractical idea and gets frustrated when others don’t immediately embrace it. But there’s a good reason for the conservative approach: codecs aren’t just judged by their theoretical performance; they have to be implementable, efficient, and compatible with real-world constraints. They also have to somehow make financial sense and cannot be given a way without some form of compensation.

It’s a bit like developing an F1 car. Or a cutting edge airplane. Lots of small optimizations that have to work together. Sometimes big new ideas emerge but those are rare.

Until the new codec comes to together all those small optimizations aren’t really worth much, so it’s a long term research project with potentially zero return on investement.

And yes, most of the small optimizations are patented, something that I’ve come to understand isnt’t viewed very favorably by most.

Are you really saying that patents are preventing people from writing the next great video codec? If it were that simple, it would’ve already happened. We’re not talking about a software project that you can just hack together, compile, and see if it works. We’re talking about rigorous performance and complexity evaluations, subjective testing, and massive coordination with hardware manufacturers—from chips to displays.

People don’t develop video codecs for fun like they do with software. And the reason is that it’s almost impossible to do without support from the industry.

That’s hardly true. Nvidia’s tech is covered by patents and licenses. Why else would it be worth 4.5 trillion dollars?

The top AI companies use very restrictive licenses.

I think it’s actually the other way around and AI industry will actually end up following the video coding industry when it comes to patents, royalties, licenses etc.