Hacker News new | past | comments | ask | show | jobs | submit login

Read this reaction from xiphmont: https://xiphmont.dreamwidth.org/84214.html

It paints a quite different picture.




This comes off as rather dismissive. Xiphmont simultaneously claims that Jarek's work is useless ("the performance claims just don't hold up") and that Google needs a defensive patent on it.

As someone on encode.ru pointed out, "if Google genuinely wanted this as a defensive patent, then the right approach is to work with Jarek and pay for him to file a patent on ANS itself blocking as many of the spin-off patents as possible." They didn't even talk to him before patenting his work.


Not useless, just not as good as Jarek liked to claim. It's a neat technique but hardly a breakthrough. In the end, for AV1, Daala_EC was better.

As for 'needed a defensive patent', need is the wrong word. In the current patent game, you pile up defensive patents to be used in aggregate as part of a MAD-style game. It sounds insane and it is-- but it's also currently the least insane option.

In any case, Google isn't technically in the wrong here. They're not patenting jarek's work, they're patenting improvements they made to it that were necessary to use it in AV1, which they dedicated engineers to do because Jarek declined to do it himself. Only after Google put in that work (and implemented multiple suggested improvements on top of it to try to bring its performance up) could we even measure the results.

But just because Google isn't wrong, doesn't mean they weren't assholes. Having negotiated with them for things in the past, they will promise the moon and the sun and the sky, and commit only to the parts that get them what they want. I can fully believe they verbally promised Jarek far more than he got.


That doesn't sound like the right approach. Google has the resources to actually defend the patent. While ideally/ethically Jarek should be the one with the patent, if the patent is only being used defensively, Jarek probably isn't the right entity to actually defend it.


Google could also buy the patent from Jarek. At minimum, they should have talked with him.


> "the performance claims just don't hold up"

see now the fatcs in https://sites.google.com/site/powturbo/entropy-coder and how rANS based entropy coder outperforms the current AOMedia AV1 entropy coder by 7 times.


Put it in AV1 and it doesn't.

The entropy coder in AV1 right now is Daala_EC. It's the entropy coder because we measured it directly against rANS and the other candidate encoders and it performed the best.

Hell, you can check out the code, build with the different entropy coders, and test it yourself.


Already done. The claims are not based on speculations, but on numbers coming from verifiable benchmarks with "https://github.com/powturbo/TurboBench" including the Daala_EC and the AOM AV1 entropy coder.


> So my action item for Jarek is: ... Find a friend with a big bag of weed, and take a few days off.

It's not just condescending, it's simply offensive.


Because Jarek is frankly out of line, and I see no reason to sugarcoat it.

Google pissed him off (and I'm sure he has reason to be pissed) but he's not exactly being truthful about what happened, or what is happening.

Being angry at Big Bad Company for a good reason does not give an engineer license to bend the truth any damn way he feels, and it certainly doesn't give license to just make shit up. He's lost some of my respect for a good reason, and if an Avengers reference offends him (or you) it's not my problem.


I see nothing substantive in the post you linked to. Can you clarify why you think it “paints a quite different picture”?


What is this? Is xiphmont a Google or Mozilla employee who works on AV1? He's wildly unprofessional, especially all the profanity and the advice to smoke a bag of weed.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: