Hacker News: MadameMinty

New comment by MadameMinty in "H.264 Is Magic (2016)"

MadameMinty — Thu, 29 Oct 2026 18:18:30 +0000

JXL decoding can still be optimized and had significantly less time for it, so let's not be so sure if that statement holds. Last update brought much improvement at just the area you mention.

And AVIF compression ratio beats JXL only conditionally. E.g. my photo library had AVIF at 29% and JXL at 25% the input size at the same perceptual quality (SSIM2=83). Sometimes it's on par. Sometimes AVIF wins. JXL wins almost always. Not to mention you can losslessly, reversibly transcode lossy JPG files to JXL, getting significant size reduction for free.

New comment by MadameMinty in "H.264 Is Magic (2016)"

MadameMinty — Thu, 29 Oct 2026 18:09:55 +0000

I hope so, but I disagree with the size of the "except": JXL performs better than APNG and GIF, and still supports transparency and lossless encode. Video codecs struggle here. So even with animation it still depends!

There's also a case for AVIF in some use cases (for this image it performs poorly). But if I had to choose one, I'd choose JXL.

New comment by MadameMinty in "H.264 Is Magic (2016)"

MadameMinty — Thu, 29 Oct 2026 16:52:47 +0000

For completeness, my background job of the highest imaginable PNG compression yielded optimized: 528313 bytes. Took only several minutes...

New comment by MadameMinty in "H.264 Is Magic (2016)"

MadameMinty — Thu, 29 Oct 2026 16:49:36 +0000

av1_nvenc exists therefore it's mainstream :)

New comment by MadameMinty in "H.264 Is Magic (2016)"

MadameMinty — Thu, 29 Oct 2026 16:46:51 +0000

Yes, it's very disheartening. We're approaching two decades of WebP, after all. At this point it's not even worth implementing its support because JXL is superior in every use case.

New comment by MadameMinty in "H.264 Is Magic (2016)"

MadameMinty — Thu, 29 Oct 2026 16:41:39 +0000

Worth pointing out that PNG is also just awful and should probably die like BMP died. WebP was already available at the time this article was written:

PNG: 1039993 bytes - baseline

optimized*: 541539

WebP: 398974 - lossless mode, of course.

JPEG XL: 310778 - lossless, quick, and easy.

*after monumental compute effort; still probably some room for improvement but the effort is ridiculous.

New comment by MadameMinty in "Gemini 4 Carbon"

MadameMinty — Wed, 28 Oct 2026 20:50:38 +0000

Not pedantic, the plural is the point. More than one star (Sol).

New comment by MadameMinty in "We should be more ambitious with the software we build"

MadameMinty — Sun, 25 Oct 2026 14:17:57 +0000

Impossible ≠ unfeasible.

It's extremely unfeasible.

New comment by MadameMinty in "We should be more ambitious with the software we build"

MadameMinty — Sun, 25 Oct 2026 13:05:04 +0000

Which is a great pastime but not a feasible way to actually make games.

New comment by MadameMinty in "Show HN: Gemini 4 Argon plays StarCraft"

MadameMinty — Fri, 16 Oct 2026 15:58:45 +0000

for AIur

New comment by MadameMinty in "Gemini 4 Argon"

MadameMinty — Tue, 13 Oct 2026 23:12:44 +0000

"You've had one, yes. But what about second Ultra?"

New comment by MadameMinty in "Show HN: Use your OpenAI subscription to create audiobooks"

MadameMinty — Tue, 13 Oct 2026 15:09:33 +0000

Behold, an audiobook: https://drive.google.com/file/d/1v4Iux5ppJe5SbE4TcpV9NQTsY0N...

New comment by MadameMinty in "Janas: Run LLMs larger than RAM by streaming experts from SSD"

MadameMinty — Tue, 13 Oct 2026 11:24:36 +0000

s/understate/overstate/ I mean, I get up to 90 tok/s on consumer hardware (16 GB VRAM + 64 GB RAM) using IQ3_S, which so far is holding up quite well.

New comment by MadameMinty in "Janas: Run LLMs larger than RAM by streaming experts from SSD"

MadameMinty — Tue, 13 Oct 2026 07:15:23 +0000

Then check out those:

https://github.com/prabanta-dev/janas <- this, focuses on Qwen3-Next-80B-A3B.

https://github.com/Niko1221/Strata <- Qwen3.8-Flash-Next. I use this (tuned) and I cannot understate how satisfied I am.

https://github.com/antirez/ds4 <- DeepSeek 4 family + more.

New comment by MadameMinty in "Beam: Reflection's 501B open-weight model"

MadameMinty — Sat, 10 Oct 2026 13:33:22 +0000

I myself am on the edge of calling it sarcasm. On one hand, I meant it. I believe it to be true. On the other, I recognize that the options are depressingly scarce and uncompetitive, and I'm saddened by it.

New comment by MadameMinty in "Beam: Reflection's 501B open-weight model"

MadameMinty — Tue, 06 Oct 2026 08:32:35 +0000

Yeah, we should only promote, like... Finnish, Norwegian, maybe Swiss products, like Apertus. Not this unethically-made baby oil[1] from repressive torment nexuses of USA or China.

[1] https://i.imgur.com/rJSG019.jpeg

New comment by MadameMinty in "OpenDLSS: A Vulkan Reimplementation of Nvidia's DLSS 5 Neural Rendering Network"

MadameMinty — Thu, 01 Oct 2026 21:55:42 +0000

Pretty sure the author was Claude

New comment by MadameMinty in "OpenDLSS: A Vulkan Reimplementation of Nvidia's DLSS 5 Neural Rendering Network"

MadameMinty — Thu, 01 Oct 2026 08:56:04 +0000

Yeah, I call it bugfix storytelling. Once upon a time this class far far away had this red hooded method...

Especially egregious if both adding and removing the thing happens in one commit. Git should be telling the story, and if it can't then there _is_ no story!

New comment by MadameMinty in "OpenDLSS: A Vulkan Reimplementation of Nvidia's DLSS 5 Neural Rendering Network"

MadameMinty — Thu, 01 Oct 2026 08:41:03 +0000

I'm more upset about it being factually wrong, e.g. both mentions of "git-ignored" are absurd (why would you mention it if it's not in the repo?) and wrong (they are in the repo).

New comment by MadameMinty in "Sonnet 5.5"

MadameMinty — Mon, 28 Sep 2026 18:24:50 +0000

That's frankly hilarious. What was the fallback for Opus 5.5? Was it Sonnet 5 or 5.5?

I suppose it also explains how FrontierCode scores seriously dip at Opus/Xhigh and Sonnet/Max?