JPEG XL

Testing GPT Image 1.5

I am the opposite, I dislike active cooling conceptually. It is literally combatting wasted energy (that goes into heat instead of compute) by wasting more energy. Also I don't like moving parts and noises.

Passive cooling ftw

how does energy go into compute and not heat?

as i understand it ~100% of the energy going into the compute parts just ends up as heat

I guess technically yes, "compute" is a bit hard to define in physics. But clearly if one processor requires 50 Wh to perform X operations and another requires only 5 Wh to perform the same X operations, then the first one is wasting at least 45 Wh on producing heat.

We should put all datacenters in the basements of apartment buildings in cold countries, so that the datacenter IS the central heating system of the building.

✨ district heating ✨

I know that not everywhere has the infrastructure for that, but being able to efficiently move around waste industrial heat for residential/office heating is very OP. Electrical and chemical energy aren't the only worthwhile energies to have large-scale distribution infrastructure for 😄

Dystopian future where you have to keep talking to the AI or else your building gets uncomfortably cold

https://torrentfreak.com/hollywood-warns-extortionary-codec-patent-fees-could-hike-streaming-subscription-prices/ Let them eat themselves.

I thought the patents for H264 expired?

I actually agree completely, I guess conceptually was not the right word (like I said, I can't really articulate it). maybe aesthetically? I enjoy there being a little fan. but I recognize passive cooling is better

technically it probably doesn't matter how energy efficient most freestanding household electronics are if it all gets turned into heat and subtracts from the energy usage of your central heating :)

if you don't ever use cooling air conditioning, that is

don't let logic get in the way of profiteering 🙂

Would be very funny if MPEG codecs end up only used by pirate groups if aggressive patent litigation forces legitimate users into AV1/2

there's also that one company claiming you need to buy a license from them to use AV1. I wonder if they're still around

I like active cooling because I need a space heater anyways.

how efficient are processors if you see them as a very fancy electrical heating device for warming up a room?

As good as any other electrical heater

I am bad at physics/thermodynamics, but I would imagine not all electrical heaters have the same efficiency in terms of converting kWh into an increase in temperature? Or are they all the same and a Watt of electricity just gets converted in a Watt of heat and that's it?

Basically, yes, they're all equally efficient

the only difference is where the heat goes

If you have a hot wire, then I suppose some infrared (and maybe even visible) light could leave through your windows, but anything that doesn't is gonna be absorbed. Sound is gonna end up as heat too eventually

speaking of heat, wiring a generator directly to oven element is a surprisingly good dump load

I can imagine, big resistor designed to handle getting really hot

electric space heaters are by definition 100% efficient (ignoring losses in the form of sound/light if they produce any) there is one exception: devices with a heat pump are *more than* 100% efficient. it turns out, that rather than turning all electricity into heat, which is 100% efficient by definition, you can use that same amount of energy to transport more heat from somewhere else into the room

so long as the outside temperature is above absolute zero (−273.15 °C), then in theory, there is heat to be taken out of the outside air and transported to the inside air in your room

most heat pumps only work to around -20 unless you get a cold weather one, and those are dumb pricy

also why no window unit <:AngryCry:805396146322145301>

this is how air conditioning units work. you have two units, one inside and one outside, that form a loop containing a certain gas, which will cool down (absorb heat) when expanding, and heat up (disperse heat) when compressing. the best part: depending on the model, these two units can *switch roles*, meaning you can choose which side expands the gas and which side compresses it. which means a single unit can be used for both cooling (stealing heat from the inside and sending it outside) or heating (stealing heat from the outside and sending it inside)

yeah unfortunately there are practical limitations, like also you can only steal so much heat from the outside before the metal fins are covered with ice, at which point they have to defrost. but I'm optimistic: as long as the temperature is above zero Kelvin, there is heat to be stolen from the outside and transported to where it's needed >:3

heat pumps are so cool

I just wish they were viable for me, 3000cad for a mini split is the cheapest available that I can find that would actually work here since I hit -26 to -30 regularly

Then there's me in Britain, where AC is basically commercial only unless you spend £30K to retrofit

wtf, no window units or mini splits?

relatedly: https://www.ewz.ch/en/business-customers/real-estate/reference-projects/lake-zurich-lake-heat-networks.html

> The thermal energy stored in the lake – including low temperatures of around 5 degrees in winter – is harnessed by efficient heat pumps. In relation to the volume of the lake as a whole, the heat extraction is so minimal that it has hardly any effect.

that's neat, I remember seeing Japan doing something similar too

They're building a huge heat pump in Copenhagen to pump heat out of treated sewage. The waste heat from the "let bacteria eat everything in it" approach means it ends up pretty lukewarm, apparently

It's one thing to abstractly know that digestion is basically just burning but happening at a lower temperature thanks to the catalytic power of enzymes, and is thus theoretically just as exothermic, but I never considered the implications of it at those sorts of scales. It's not like bacteria cook themselves when growing in the lab, after all. OTOH, a living human gives off around 80 W of heat

I just want window unit heat pumps that can work at -28c is that really to much to ask for T.T most only work at 5f or above

Our windows open horizontally and usually have limiters, so window units wouldn't fit

that's a good way to reduce waste though you'd still need an active cooling system for that, and probably AC cooling in the summer.

the Internet Archive does something similar, in their main location, the server exhaust is used to heat the building

ours does too, we just use a lot of styrofoam

no limiters tho

https://www.tomshardware.com/tech-industry/nearly-7000-of-the-worlds-data-centers-are-built-in-the-wrong-climate

This does assume that the servers are mostly/mainly air-cooled, directly with outside air? For water cooling, humidity management management doesn't really matter because the cooling happens outside in cooling towers anyways

Or at least not any more than for any well air-conditioned building anyways

probably even with air cooling with the tiniest amount of mixing

in such low temperatures, “absolute” humidity is low even when relative humidity is high

You do want to keep the relative humidity at around 40-60% because of ESD, so at low temperatures it certainly is something that needs active maintaining if you have a purely "air replacement" system (which would be about most efficient system for purely air cooling)

but if your effective heat-sinks are outside, you don't need that much air replacement inside the building besides what's needed for people inside to breathe, which becomes way, way easier to handle

I got a book about the history of paleontology which has a bunch of nice illustrations, but when I quickly looked through it I was a bit shocked when I saw some bad DCT artifacts and resolutions so low I could make out individual pixels. the majority of illustrations are fine, but it's *weird* seeing compression artifacts in print

I guess when you're covering such a broad topic and need a bunch of illustrations, you're bound to run into situations where there just isn't any better imagery publicly available, and not every museum is going to go out of their way to digitize a specific specimen tucked away deep in their archives just for you

A book, sold for money, using random freely available illustrations rather than bespoke or at least licensed professionally made paleoart seems very odd to me.

I didn't mean freely available, I meant available to the public (and by extension the author) at all. there is a section near the end that has sources for all images. it seems to be a mix of licensed copyrighted, public domain, and creative commons licensed works

the book is a translation of *Paleontology: An Illustrated History*, ISBN 9780691220925 in case you're curious 🙂

ah

https://www.reuters.com/business/media-telecom/softbank-races-fulfill-225-billion-funding-commitment-openai-by-year-end-sources-2025-12-19/

Stole this from <@245794734788837387> https://twitter.com/dreamboum/status/2002882752094654754

Neat

AI slop

I'm not orange 😤

https://www.youtube.com/watch?v=0ZVFAkdRMRY

>Arm >Reduced instruction set computer >Can execute `char foo = *(ptr += 5)` in a single instruction (x86 can't do this) Anyway I'm learning aarch64 assembly as a precursor to learning NEON

Meanwhile I was kinda thrown into it all because of the project I contribute to these days primarily.

https://github.com/zampierilucas/scx_horoscope

Merry <:JXL:805850130203934781>mas

merry christmas! do we have a JXL variant of the "share to spread 2 KB of christmas joy" but it's 300 bytes

I tried to make one but it was unrecognizable, you could probably get there with JXL art

oh we have this

You could do something like this: https://discord.com/channels/794206087879852103/824000991891554375/1321246465993212036

.

Drats, mobile navigation was too slow haha

Would be nice if someone could draw that christmas tree and that little guy and the text "repost to spread 1KB of Christmas jxl" using splines, and do it in around 1KB 🙂

Show how good 1KB could be

verywaiting 300000 GB spotify drop <:banding:804346788982030337>

Found my old Eee PC 901. Gonna gut it put an Arm SBC in it.

ready to forward to other servers to spread the word of jpeg xl

uh I mean spread the christmas spirit of course

it's a good thing discord puts the proof of the size right below the file name. I've seen that "2 KB of christmas joy" meme a bunch of times and they're usually PNGs of screenshots with considerable generation loss, that are themselves like 2 MB

that PNG render of the JXL is so large 😭 I tried lossless WebP but it only reduced it by like less then a megabyte so I thought "f it let's see what lossy looks like" so uh here's that I guess

I went for some rather funky red and green (christmas colors) tablecloth kind of thing with some pattern created with Squeeze that will probably be tricky to compress for any other codec, lossy or lossless, which I kind of did on purpose to make the gap in file size extra large 😛

someone on another server thought it was fabric, so it worked :)

the photon noise alone probably removes any chance for other codecs to try to get smart with MA-tree-like features

Ah right the photon noise is doing a lot of heavy lifting there too.

You can also make a shiny, somewhat blinding HDR version: https://jxl-art.lucaversari.it/?zcode=3VXJihRBEL3HV8QP9BBLZkbkxYMgeHIbxPMwXToFY0_rlBe_3ldLu4C3EgakKOLFWpn5XlLPx-k4nKc7Tno33JqK8Ju39PrrOJymm2l8OLHSq4fxcWC5cry2vTMuG1YW-jAeMUS7Cb0cxk93E6uk0Pvz483n8_2AKXR9vh9PA-lVZdnz_IMJZWe_8QFbr_t2UTq7NTJT9gLbO6tX8iLs4uRd4DuVGuy1UNPOFg3W2LxR9cKW8M2XfGA9VoNCYB1-gy2dcqmf84bvOUViflXKbKgvlIE6C-oF84pSxzrMke-NXSv1hnWWJPDZl0KVnogUUk2MLEaKMFJLpHPpANJyBT6PwTZenI6bAH4pYTcLT9s_34GdGqjGmUqawVkqGThTsTngANXI3birUIvkGgl2YbWBxcY1wV5x-PWn36CWYqsKqgZUUjmQr2A_pGGeIK4UrtzNKGphME9ZOwdUMrOeswoW30GkBnfIUlXXCthEK3SgrXB6rGQnJKNaZ4AuE11rPDEfqgOoa41BZ2l9BoGBNkukcGDTf5GIPfFF39_v-wTScFutbCC6_HZI_8kRJdv-I-pyASZ_6ujLt2H4PtD4kW_5GdsFJB34epj4YBuweknJBeiWks3GBeAvSfQD

https://embed.moe/https://cdn.discordapp.com/attachments/806898911091753051/1453885279357108264/Merry-jXl-mas-HDR.jxl?ex=694f13ae&is=694dc22e&hm=19574da05d0a380dae60346359b924b4690be22c17ee959312774978aff5f1df&

I can't set the noise exactly to zero because it fails to encode, so one of the parameters is set to 0.01 and all the others to 0. it looks like without the noise PNG would have a shot thanks to LZ77 unfortunately

this is without noise: https://jxl-art.lucaversari.it/?zcode=3VXJihRBEL3HV8QPTBNLRkbkxYMgeBMcxPMwXToFM22r5cWv91UvLuCthAGpQ7xYKzPfS_LlvOyn4_LARW-ne1MRevNlng7L3TJ_OrASvZ_3SOswodfT_PFhYZUSenf8evd0fJzWktvj43yYSHfBsuX7BxPaxn7jG9lZbNtFG-zWyUzZG-wYrB7kTdjFyYfAd2qR7NGo62DLDmts3im8sRV881M-sR6LpBRYh99h26A61a95w_-csjA_lKo66htVos6SRsO8pjSwDnPkR2fXoNGxzlYEPsepUGUUIo1UCyObkSKM1CkyuA0A6XUGvo7BNl4d9hcB_FLCZhaet99Zdhs1EMZVSlrJ1YIMnKnYGnCAMHI3HirUsziywC6sdrDYOQrsNYcfP_0OtTQ7qyA0oZLgRD7AfkrHPEFcKV15mFFGYzBPFYMTKllZr1UFJ99BpCYPyFJVzxWwhVboQHvj8jyTXZCMaqwAXSZ6rvHCfKgOIM41Bp2VjRUkBtoqkcaJTf9FIvbMF317v28TSMdttXYBOeS3Q_pPjqjYth_RkCsw-VNHn79N0_eJ5g98zy_YrqDohm-nhW_sAiyuKbkCvaTkYvMK1vePfgA

wow that PNG is a lot larger than I expected. I wonder if the image is too wide for LZ77 to back reference two lines up (to make up for the lack of a NN filter which seems to be the best shot other predicting formats would have for this image)

encoding the no-noise version directly from pixel data at high effort resulted in a 307 KB JXL, not bad!

That's only an increase of 1000x

lol I meant not bad compared to the PNG

JXLmas

How do u ppl find old pcs

Or like netbooks

I assume through secondhand marketplaces such as ebay or facebook marketplace

I’ve checked those places <:galaxybrain:821831336372338729>

Thanks PayPal, you absolute sods

Closed it within 12 hours. So much for a January 7th investigation time

I charged back and never went back to paypal

https://www.wheresyoured.at/the-enshittifinancial-crisis/ Ed Zitron's last free article for this year.

A lot of people find old computers at electronics recycling places I found a good one at a university surplus store

For laptops, old Chromebooks on eBay are usually a good option, Linux works very well on them

Although you might need this https://github.com/WeirdTreeThing/chromebook-linux-audio

Thxx ill check that out!

https://thejenkinscomic.wordpress.com/2025/03/10/full-glass-of-wine/

Specifically with wine, I do tend to agree with A.I. WIFE that a glass that is 1/3 wine, 2/3 air is actually a perfectly full glass of wine. Wine glasses are not like water or beer glasses, they are not supposed to be filled to the brim.

Happy New Year!

happy new year! 🎉 😊

Happy new year :)

Happy new year!

Happy New Year!

naturally the netherlands temporarily turned into a warzone

HNY++

https://x.com/i/status/2008117010543382776

Funny domain name

https://www.youtube.com/watch?v=ZWQ0591PAxM

https://fxtwitter.com/vxunderground/status/2008229485993488549

Nobody knows what that Copilot is which Copilot

Enron speedrun.

people are going to be asking Copilot how to disable Copilot in Microsoft 365 Copilot

And it will answer "Sorry, I don't know how to do that"

"please purchase a subscription to disable Copilot" <:NotLikeThis:805132742819053610>

the next stage of enshittification: instead of having to pay for features, you get a bunch of features that only investors care about and you have to pay to turn them off <:KekDog:805390049033191445>

I double checked this and it seems it's just called Microsoft 365 <https://www.microsoft.com/en-us/microsoft-365>

https://www.office.com/

the screenshot comes from this page

ahh

yeah I also found this just now [The Microsoft 365 app transition to the Microsoft 365 Copilot app](<https://support.microsoft.com/en-us/office/the-microsoft-365-app-transition-to-the-microsoft-365-copilot-app-22eac811-08d6-4df3-92dd-77f193e354a5>)

save yourselves, use libreoffice or onlyoffice

and make sure onedrive hasnt secretly uploaded your confidential docs and deleted the local original

use typst!

I use the subcription of Microsoft 365 without Copilot

2026 meh just use 2007

wish older office verisons worked on new macs

The dumb M chips dont let u download shittt

:((

Ik i could vm it but thats such a procedure aughhhh

https://wccftech.com/ai-cost-increase-allegedly-behind-microsofts-rumored-new-massive-round-of-job-cuts/ Even the Azure team could be affected, all because of Slopya Nadella. Self-cannibalism.

this bubble isn't bursting as quickly as I'd like

Microsoft is so poor with hundreds of billions of dollars

we even got reddit spam now smh

<@794205442175402004> might want to remove that latest reddit post, given the kind of spambots we've had here lately...

did you know that while the canals between the eyes and the nose have a valve so that it only goes one way (eye -> nose, hence the runny nose when crying), that valve is not perfect, and if e.g. you blow your nose too strongly, air can come out of your eye?

I didn’t know until roughly 15 minutes ago

now I do

I discovered that roughly 20 years ago

https://www.theverge.com/ai-artificial-intelligence/857640/openai-launches-chatgpt-health-connect-medical-records Oh my god, what the fuck?

great timing, I was just watching this that someone shared https://www.youtube.com/watch?v=VRjgNgJms3Q

oh, no great timing, you sent that hours ago

this video explores real cases of "AI psychosis" (where some people become delusional after talking to services like ChatGPT) by going along with what it says and seeing how far it goes

Scary

https://vxtwitter.com/RobertFreundLaw/status/2006111090539687956

Tbh OpenAI should be on the hook for absolutely everything ChatGPT says

Is Google's propaganda getting spread here?

Now everywhere online is praising Google as GOD

?

I don't know about you but I always pray in the direction of the nearest Google datacenter

Google should be on the hook for aboslutely everything Gemini says :)

I'm not sure if it makes sense to expect LLMs to say sensible stuff or to make companies responsible for what comes out of them. I think people should just learn to understand that these things are just super powerful hallucinating autocomplete systems that may sometimes be useful but shouldn't be trusted at all. If someone kills themselves or someone else because the LLM said so, I think that's basically the same as someone who does that because they played too much GTA or saw too many violent movies. Like when movie theaters were a brand new thing, people would get confused and think that what they see is real, and when a train comes heading towards the camera it would cause a panic in the public. There's a difference between fiction and reality. We don't make it illegal to have movies where people get murdered, and we don't sue the actor or the movie company when they are depicting something that would be a horrendous crime if it were real. Of course there's some room for debate about where freedom of speech/expression ends and where incitement to crime starts. Making an action movie containing lots of violence is considered fine and just fiction, but if you'd make a propaganda movie with the goal of convincing people of nazism and to kill people because they belong to some 'unwanted' groups, then this would/should not be considered fine. I think something similar is true for AI: the intention of the designer matter, if you try to make an LLM with the goal of manipulating people, or if you use deepfakes to mislead or create harmful media, then definitely you should be on the hook. But making this kind of technology watertight is impossible, and I think it's basically the same as requiring word processors to not allow people to write harmful text.

That said, I do think AI companies could do more to avoid giving the impression that their bots are human and sane. Especially those "AI companion" companies are basically actively trying to mislead people into antropomorphizing LLMs.

> mislead people into anthropomorphize LLMs if company are honest to admit LLM is nothing more than interactive library, how they can scam investors? "trust me bro, LLM will be AGI, robots will take over every job, human will live in heaven in the earth"

anthropomorphize LLMs, yeah I make them consider themselves as furries

only Copilot refuses

I’m not sure they have that much to do with one another? does giving the impression that LLMs will take over jobs really require them to speak like this: https://www.threads.com/@vanilla_lea_sims/post/DTPlmhujbh8 https://www.threads.com/@moonlight.bzh/post/DTNsAPvCKCN

(this, by the way, is one reason why I prefer Gemini over ChatGPT; it seems to have less of a tendency to do that)

(I ask it a question, it answers the question, done, no pretense)

https://www.myabandonware.com/game/kkrieger-chapter-1-cl1#download

Less than 100kb 3D shooter

The other day I wrote a hello world program in raw assembly and it turned into a 70 KiB executable

If someone wants to test this on windows 11, Enable program compatibility to windows 7

They say meh to JXL but yes to websites running neural networks on your hardware... https://github.com/mozilla/standards-positions/issues/1215#issuecomment-3520278819

I guess it's better than webgpu if it can use the npu? Note that Firefox still doesn't consistently enable webgpu by default...

that's a yikes

even better:

here is the commit, by the way: https://github.com/mozilla/bedrock/commit/d459addab846d8144b61939b7f4310eb80c5470e

iirc this was because "selling your data" was not legally defensible under a certain US law. iirc, even having sponsored shortcuts or search engine deals somehow counted as selling your data. so they didn't remove this because they are actually selling your data

on an unrelated note, I've seen Theo's video about JPEG XL and it looked like there was basically no research done, add to that that they didn't seem to know a lot about image compression, and it made for an overall really poor video. if that's the standard they strive for, I wouldn't trust them to report other things correctly either

Is this what they call Waterfox?

lol is this due to the bug where chromium swaps the color channels of JXLs for some reason?

Yeah the Android bug

so is a Long-term nonbinary love relationship XL XL?

I need me some of that

Another hit for translators https://blog.google/innovation-and-ai/technology/developers-tools/translategemma/

https://youtu.be/ORffRIkY7O0

classic album made a bit more headphones-friendly

(standard stereo mix for reference: https://youtu.be/IirnyKNJNiA)

now chatgpt.com are broken in Firefox, only Chrome can visit

I can visit with Safari

> Fixed compatibility problems with websites that use the new Compression Dictionaries technology, such as ChatGPT, by temporarily disabling the feature. (Bug 2010712) > https://www.firefox.com/en-US/firefox/147.0.1/releasenotes

<:tfw:843857104439607327>

ah, shared brotli

so ChatGPT uses that? nice

that's where the webserver and browser negotiate a dictionary beforehand, right? pretty neat, I knew that was possible but didn't know of any sites that used it

google search does https://developer.chrome.com/blog/search-compression-dictionaries

https://bsky.app/profile/edzitron.com/post/3mcknjfom3c26

rogers the compnay me and my brother are with for our phones, cancled my order for the pixel 10 pro xl, black frida deal for 400 down, 0 dollars a month, because they no longer have the jade (green) version. now to re-order the phone they want 1100 down and 20 dollars a month... their customer support is about to have a really bad day on monday

"4chan is full of nice people who don't hate anyone" t. Someone who definitely knows stuff

4chan is filled with a larger spectrum of people than what people claim

this is really cool

related, I remember Discord experimented with custom Zstandard dictionaries for a certain type of small message that they send to clients at a high rate, but iirc dictionaries had marginal improvements and it wasn't worth the extra complexity for them. that makes me wonder if they didn't train it properly

https://discord.com/blog/how-discord-reduced-websocket-traffic-by-40-percent

idk where to put this, but id like to report this user who randomly started messaging me, and the second i accuse him of being a scammer he blocks me

These scammers r so stupid

Convinced some r bots

come on man if youre gonna scam you might as well try something that hasnt been done a billion times before

<@&807636211489177661> they're still here: <@1461965822921543794>

https://www.reddit.com/r/youtube/comments/1qdvgmc/uploading_srv3_subtitles_got_brokendisabled

what a shitty company

Copying my messages from <#822120855449894942> for when it's all deleted Somehow it's getting worse every time... Turns out I have permission to kick the spammers, but can't delete the messages or ban them permanently, so I can at least stop them from hitting every channel in future

Can tell it's a real person and not just a bot, because after I pinged for admins they started doing a one-line message on every channel they could until I kicked them

I know it's early in the morning, but doing an extra <@&807636211489177661> to try and clean up <#822120855449894942> ASAP. Should still be able to ban the user even though I kicked them

Banned this one but yeah I should figure out how to hand out more banhammers

Ok everyone with dev role can now also ban and timeout, and delete their spam

great, my timezone doesn't overlap much with others so that's good to have

These spammers are not very clever if they think anyone on a server like this will be falling for their scams

Reverse spearfishing

My sleep schedule is as consistent as Chrome's opinion of JXL, so I'll probably catch a lot of it

I do wonder why they have been coming into this server. my theory is that something is seeing "JPEG" (as in images) and "X"/"XL" (as in NSFW) and making some leap in logic but idk

Probably just randomly any server they can join

The current invite link is definitely on a list now, we've had significantly more in the past month than the past year

are they even using a invite link? because IIRC this is a public community server so it ends up getting pushed to the Discord discover section

Invite link would be the easiest way for them to automate joining

I know Discord used to allow seeing how users joined the server, or in the manage links section it would display the users who used it

You raise a point, it's all been NSFW 'offers' as opposed to usual DM spam

what they don't provide audits for invite links?

Last time I tried they did, but it was years ago and I think needs more permissions to see than we have

Hmm, looks like it got removed. Invite links only display a number now, and the Audit log only shows bans/kicks not joins

Depending on how big of an issue it becomes, we could always enable the Onboarding feature (or at least I think that's the name) Then new members get a pending status instead of immediately joining, and we'd accept or reject based on their profile

I've seen some servers setup a text channel where if a account posts in it they get banned and have all their recent messages removed

Interesting approach.

I think that could lead to random bans if people come here just wanting to ask a quick question or something

And if the name was obvious, they'd just skip it. It was a person this time who reacted to me pinging the admins

also core dev? I don’t seem to have that permission

I wonder if the server is being targeted in some form then? could be the invite link getting added to some illegal advertising list or something? question: what was the reaction added to your message and was your message sent as a reply message or not?

might be worth it to try creating a new permanent invite link, deactivate the old one and replace the relevant places with the new invite link such as the website and the reddit and wherever else to see if that stops it

I assumed that would be automatic, but looks like it wasn't. Now core dev too.

I would expect the bots are just using lists of publicly indexed Discord servers. The YouTube user No Text To Speech posted a video on the history of the Discord CSAM problem a little over a month ago: <https://www.youtube.com/watch?v=NTY0d9KM0Hw>

I would be a bit careful in the future handing out permissions. It's not that I don't trust the devs and core devs it's just that it takes one unfortunate series of events to lose a lot, especially with deletion related permissions.

It wasn't a literal reaction, but the first few channels were big blocks of formatted text with images, once I pinged the admin role they started copy pasting a one-line message until I kicked them

There was a minute or two between each message, up until the one-line spam

Presumably the bot trying to evade slow mode timeouts or some kind of flood control.

We also don't directly have deletion permissions, just the prompt that happens when you ban someone (I assume, maybe it doesn't prompt because we don't have the perms)

This wouldn't protect against someone banning every user and setting it to delete all messages instead of recent ones. If only Discord had more fine grained permissions to allow giving people the ability to only delete *recent* messages and not all. It's always made me uneasy that to allow someone to delete messages allows them to delete **all** messages.

Let's just not do that

I theoretically can delete this server but I will not do that 🙂

Plus Discord has a permissions hierarchy that prevents some things, e.g. people with the dev role can't ban other dev or core dev members. Of course, Discord doesn't clearly document the permissions system (or if they do I can't find it).

I would expect that devs not having the option to delete messages in the chat would carry over to the ban option, but I obviously wouldn't be testing that.

I should backup the server again

Permission system of Discord is messy as permission system of Windows<:ugly:805106754668068868>

do we do this already? https://support.discord.com/hc/en-us/articles/4421269296535-AutoMod-FAQ#h_01GV3GQSB85NVWN1KQJ8AKZYC8

either there's something about this server specifically that sets them off or there's some difference in the automatic moderation because I am in many mid-size public servers and none have a spam problem that's this frequent

https://fxtwitter.com/beneater/status/2012988790709928305

same

Making rice wine. My first time making any kind of alcohol.

Watch out for methanol

interesting first choice, not a bad one, just interesting

The risk from methanol, fortunately, is approximately 0.

Well, I'm on a koji kick :D I've got miso fermenting, gonna make mirin, some variation of soy sauce, pseudo-tempeh with koji instead of tempeh starter.

hope it turns out well

Rice wine is available in almost every shop here

a basic flavour like sauce

I really hate how absolutely every image viewer does filtering in a nonlinear color space

This is a contrived example but honestly it's distractingly visible on anything with any pixel perfect lines

They just suddenly become darker whenever you zoom in

Also the moire effect is usually much less visible if they were filtering in a linear space iirc, and if they were also supersampling then it would be completely gone

here's the image if you want to see for yourself

well really supersampling would make the moire effect always be gone but the colors would also look different than if you were viewing the image pixel-perfectly

i mean as long as the supersample filtering is in linear the interpolation can be whatever and it will look the same at any scale (just not necessarily the same as it originally looked)

from left to right: supersampled srgb interp., supersampled linear interp., 1 sample srgb interp., 1 sample linear interp.

oh whoops forgot to turn blender's tonemapping off but you get the idea

unrelated but do you guys have any ideas for this person, jpeg decoding is their main bottleneck others have suggested decoding the frame with quick sync hw decoder first, apparently the camera can't provide uncompressed

yeah jxl resampling treats it the srgb way...

(cjxl -e 7 -d 1 --resampling 2 .\canvas.png .\resampling.jxl)

jxl without resampling really messes with the colors at higher distances (cjxl -e 7 -d 5 .\canvas.png .\noresampling.jxl)

this is probably the worst case real-world image i have and with jxl resampling the problem is definitely noticable if you flip between them, but otherwise overlookable

the most noticeable places are any region with one pixel black or white lines, which get significantly darker

HF color quantization is what we've been discussing the past few months, the tables are far too harsh on the high end that the checkerboard is probably using

I vaguely remember wanting to switch it to linear downsampling, but either forgot or didn't know how. The upsampling is always the original colorspace though IIRC

nonlinear image -> to linear -> downsample -> to nonlinear -> nonlinear downscaled image should be fairly simple to change that, idk how much of a problem the upscaling would still be though

upsampling is a spec feature, and it will happen in gamma compressed XYB space if the image is XYB encoded

We could still do downsampling in linear space though, that's an encoder choice.

I have the impression that downsampling works better in linear space but upsampling in non-linear space, and it’s not completely clear to me why

(nor whether my impression is completely accurate)

Upsampling in a nonlinear space is very similar to interpolation in a nonlinear space, which comes with some of those darkening artifacts but maybe looks a little more perceptually smooth when zoomed in

The comparison I was using was blender Eevee vs cycles, because cycles interpolates nonlinearly and Eevee interpolates linearly, you could compare some images there to see what you think looks better Both supersample linearly Just make sure that you turn off temporal reproduction and supersampling if desired and are aware of the fact that Eevee switches to a mipmap when you zoom out further

Eye of GNOME has a "smooth images when zoomed out" option which form my experience removes these types of artifacts

I'm pretty sure it does some sort of supersampling if you enable that

though idk about "filtering in a nonlinear color space". I just know whatever it's doing prevents moire effects

That's mipmapping probably

(downsampling to prevent aliasing by keeping texels closer to the size of a pixel)

Eevee does that eventually and the aliasing diminishes significantly with even just 1 sample at the cost of some sharpness (you can see a noticeable loss of sharpness even in the supersampled result)

Just turned my computer on and there's a dead pixel on my monitor 😔

https://www.jscreenfix.com/

It's not stuck, it's dead. No backlight on a black screen

After a suggestion by <@384009621519597581> in DMs, I'm taking a look at the server's member list and banning some suspicious accounts. Notably ones with advertisements in the bio, joining on the same day as creating an account, and duplicate names with only differential numbers

Nearly all have a 5 digit number at the end of the account name along with joining upon creation, but I thought I'd mention it here

I'm trying to think if there would be any logical reason for that naming pattern besides them being fake or bot accounts

Ah yes, very inconspicuous

like if you don't have a Discord account and create one on the spot does it it push people towards this sort of naming scheme?

Ah ha, Discord docs say the 5 digits are because the accounts didn't verify their emails within 7 days of creation, but it also says users need to pick a name... So still a wall of bots

in a pretty large server that I moderate, we have a 'quarantined' role. if you have the role, you can only see one channel, the #waiting-room, which moderators can also see. the channel description reads "if you are a real person and can see this, ask a moderator for help" we have some heuristics for when new accounts should get that role. I'm pretty sure it's a bot that does it, though you could maybe do something similar with discord's built-in automod rules

accounts without a verified email can, to my knowledge, not send messages at all. so I'm not sure what their plan is here. they can't do anything with these

Also, join method is `Unknown` for all of them, so I assume that's from being a Community server

After years of writing in Java/C#, I'm now trying to write in C++. Trying to keep track of object lifetimes is a pain. Using `new` is dangerous because of memory leaks, but using local variables can result in dangling pointers after they get out of scope, which is even more dangerous. I'm supposedly supposed to use smart pointers, but none of the libraries do. And library documentation rarely mention if object they accept is deep-copied, or lib takes ownership of it, or I'm supposed to manually dispose of it.

Delphi/ObjectPascal didn't destroy objects when out of scope, so I'm used to that behaviour.

write Rust 😛

https://drive.google.com/file/d/1Uj9QHE7JOzJE0OclXpdBLdYckix2bRVW/view After four years, everyone has just given up on the neural network audio codec. I made a portable version, "unpack and use" the program. No one has done that in four years, and there's no player at all. Who the hell needs Docker and containers when you can make completely portable Python programs that you can run and move anywhere you want mehhh It's disgusting that no one does like this. https://github.com/facebookresearch/encodec and I still had to change part of the source code so that it would load the model from the folder instead of downloading it from somewhere else so annoying

maybe <#805176455658733570> but I don't know example here https://discord.com/channels/794206087879852103/805176455658733570/1466072641000443964

which libraries are those? my experience hasn’t been _that_ dreadful

also, for those that take raw pointers, but at least document that they take ownership, it can still make sense to use `std::unique_pointer` locally but then `.release()` it when you pass it

(https://en.cppreference.com/w/cpp/memory/unique_ptr/release.html)

If someone here is interested in handheld linux computers for weird use cases, "Mecha Comet" campaign has started recently on kicstarter https://www.kickstarter.com/projects/mecha-systems/mecha-comet-modular-linux-handheld-computer/

Reddit finally supports auto Light/Dark mode switch

https://youtu.be/C6mbya4UU8w?t=1m50s I don’t think this video used to be 50fps, and it also looks more oversharpened than I remembered it being

has YouTube started “enhancing” videos post upload?

maybe I’m just imagining it

for me it says this when it's upscaled from the original. it doesn't say something like that for yours so I'm guessing not, but you never know with YouTube

Yes. The drama already simmered down, so YT boldly continues ahead.

https://youtu.be/s2lo5ZpOqFQ?t=58m10s the amount of banding here is rough

original looked more like this

not perfect either, but better

I most prominently notice color banding on Veritasium videos, which often feature gray gradients as a background behind graphics

If you want horrible banding, just upload a photo to Discord from mobile (Maybe just Android, not sure)

https://old.reddit.com/r/isthisAI/comments/1qps3zh/ice_fishing_photo_my_husband_took_the_dog_ice/ 😂

> The lighting is so weird that it absolutely looks like he just added a handful of png fish

It looks like a slide in a powerpoint

they turned shadows off to improve performance

<@238552565619359744> happy birthday! 🎂 <:JXL:805850130203934781>

Oddly enough I'll be wearing my <:JXL:805850130203934781> badge into London tomorrow with some friends

this one? https://discord.com/channels/794206087879852103/804324493420920833/1285254092406521876

Yeah

I got myself a new GPU. Finally I can allow AV1 on youtube without my CPU being eaten alive.

Nice!

just found out your pfp is animated, I love it lol

Dragons are pretty lit

is there an avif server?

there's an av1 discord: https://discord.gg/R6BzSQEy

yeah, there's an image channel for avif stuff

Made by a friend years ago, but I've only used it a few times when I've been gifted Nitro. Runs out tomorrow though

EnCodec Player potrable i made https://drive.google.com/file/d/1tFS2USco5U_mV03-mBpJPXloiZybkfOq/view?usp=sharing

The archive includes a 9-hour music file size is 47 mb<:Stonks:806137886726553651>

The jpeg industrial complex wants you to believe that your images are not compressed enough! Resist their oppressive regime by storing all of your images in TIFF!

Default UE5 = mainline svt-av1 8-bit. Default tweaked UE5 = mainline tuned svt-av1 10-bit. svt-av1-essential = tweaked UE5 with removed Lumen and Nanite with additions for optimized RT (10-bit MDS) and proper TAA (scene detection). svt-av1-psyex/hdr = forked UE5 made really good. 5fish = forked UE4 with UE5 features that are actually good.

you can use jpeg as a payload codec in TIFF 😉

Nooo! It's too late!

Soon they will replace TIFF with JPEG

https://www.youtube.com/watch?v=7_qhUidegUM

coming along well

jxl payload when

It's already there, DNG is TIFF so... In TIFF, adding a new payload codec is just a matter of assigning a new unique 16-bit number to use as compression_method 🙂

Oh how I wish they used the bayer channel type though...

Is there a library that can encode a JXL payload in TIFF? libvips does not support it yet, I think the same goes for imagemagick

I think GDAL can do that (for GeoTIFF), and I assume there should be something that produces DNG 1.7 which is also JXL-in-TIFF. For the "generic" TIFF, I don't know if anything exists that does that. What would be the advantage of adding another wrapper? If it's just for metadata, the jxl file format already supports Exif, which is basically the same thing as a TIFF-without-payload.

yo dawg, I heard you like formats...

JXL in HEIF in TIFF

image container quine <:Hypers:808826266060193874>

Oh good https://x.com/i/status/2019217305511567537

Only for windows? No repo?

lol

With this question it seems you have problems using Windows

Well, I don't use it

I hate source code as an ideology, I'm only interested in finished products, and this is what I made, a free, finished product, in an archive, in a folder, in run.bat You can try running it through Wine on Linux. or take the Encodec folder from the packages where the project code with the code I modified is located, and try installing all the python packages on Linux replacing encodec with my version, but this will no longer be a portable solution. Moreover, it is possible to make a similar portable package for Linux, but it is not supplied as a standard python in any way, since such a solution is directly opposed to what Linux is and what ideologies it promotes.

What are you talking about

https://discord.com/channels/794206087879852103/806898911091753051/1467955061320843348

the last part is just wrong

Hmmmmm Well, Linux is positioned as an apt/pacman like system where you type something and get a ready-made program, pre-packaged or compiled on the fly. Regardless of the presence of an executable file or installation packages. What I created—a portable Python program with fully pre-installed packages—is essentially neither Docker nor vnev nor some Conda like. What I created isn't typically done on systems like Linux. However, I want to note that I used Windows Embeddable Python, which is a portable, ready-to-use Python package that's officially unavailable on Linux, because using Python as a fully isolated software isn't common on such a system. But Which isn't particularly strange on Windows, where I used Python instead of an executable program, essentially attaching the entire Python to my code with libraries instead of - contradicts the idea of ​​Linux when all libraries and installed programs are in the system as dependencies, you just need to install them and not duplicate them hundreds of times, but this is not my way

https://ep2024.europython.eu/session/shipping-ready-to-run-python-apps-without-the-need-to-install-python/ Looks like there is this for Python

AppImage, Flatpak

Tbf I prefer installing things via package managers

funky

android is fairly antithesis to this, but AppImages

I'm also kinda iffy on package managers (including flathub and snapcraft). They tend to arbitrarily remove apps just because of hearsay. Which is disgusting and goes against everything FOSS stands for. I'll rather provide *.deb or *.appimage or something than to rely on those people's mood.

You can provide your own repo tho

But yeah AppImage is easier

deb would still require converting to other formats but doable for the interested

Funny how there is wxWidgets for iOS and OS/2 even but not Android

<https://en.wikipedia.org/wiki/List_of_platform-independent_GUI_libraries>

I don't think Android is even worth mentioning when talking about computer systems. To me, it's a joke, requiring literally nine circles of hell to create an app. So many conditions for a program to run—no, I don't take this OS seriously.

odd; their site doesn’t seem to mention iOS anywhere

and notably not on https://wiki.wxwidgets.org/Supported_Platforms

ah, there it is: https://docs.wxwidgets.org/3.2/page_port.html#page_port_wxiOS

> It is very basic in it current form, but is included for further improvements and very simple applications.

I wonder which cross-platform toolkit supports the lowest iOS version

Lowest needed to support all iOS device seems to be 3.1.3

to the extent that such devices are still around, though, there will be further limitations (e.g. the first iPhone had 128MB of RAM)

afaik, iOS 15 is enough for all currently-supported devices

which goes at least as far back as the iPhone 6S (announced September 2015)

correction: a minor update to iOS 12 was released two weeks ago

so one should target iOS 12 if they also want the iPhone 6 (non-S)

(announced September 2014)

Which doesn't help when the newest device I have is a 4S

Apparently wxiOS does support iOS 9

But I wonder if something can also support lower versions

5 or 6 would be interesting for the first iPad. (You can inofficially get it to 6)

I suspect one of the best bets might be to make a PWA

Gemini’s suggestions:

but it also mentions that the bigger problem might be: > To compile an app for iOS 5, you generally need Xcode 6 or older. Xcode 6 does not run on modern macOS (Sonoma/Ventura). You will likely need a Virtual Machine running OS X Mavericks (10.9) or Yosemite (10.10) to actually compile the code for these toolkits.

Well, you always need something something MacOS*

(*unless for simple apps)

right, but here, in addition, it needs to be old macOS specifically

Cocos2d-x 3.17 seems to support iOS 8.0+ and Android 3.0+. Latter is surprising

Apparently there is a 4.0 (https://www.cocos.com/en/post/cocos2d-x-v4-0-is-here) but https://www.cocos.com/en/cocos2d-x speaks about 3.17?!

its just an example, there are other examples too like AppImage and flatpak

Cordova 8.0.0 requires iOS 13. Adobe AIR... well not sure but it looks like they supported 32 bit devices till AIR Runtime 33.1.1.698: <https://airsdk.dev/news/2021/12/02/new-air-release>

They have some more text in PDF release notes: <https://airsdk.harman.com/api/versions/51.2.2.3/release-notes/Release_Notes_AIR_SDK_51.2.2.pdf>

»8.1 32-bit vs 64-bit For deployment of AIR apps on iOS devices, the AIR Developer Tool will use the provided tools to extract the ActionScript Byte Code from the SWF files, and compile this into machine code that is then linked with the AIR runtime and embedded into the IPA file. The process of ahead-of-time compilation depends upon a utility that has to run with the same processor address size as the target architecture: hence to generate a 32-bit output file, it needs to run a 32-bit compilation process. This causes a problem on MacOS Catalina where 32-bit binaries will not run. Additionally, due to the generation of stub files from the iPhone SDK that are used in the linking process – which are created in a similar, platform-specific way – it is not possible to create armv7- based stub files when using Catalina or later. From release 33.1.1.620, the stub files are based on iOS15 and are purely 64-bit. This means that no 32-bit IPAs can be generated, even when running on older macOS versions or on Windows.«

Ah, minimum target iOS 12 for that version

Also Android API Level 21 aka. 5.0

Whyever the latter is the case

et tu, mario 😭

hey, it’s quite convenient 😁

https://www.bleepingcomputer.com/news/security/zendesk-spam-wave-returns-floods-users-with-activate-account-emails/

anyone else get hit by this? there's another wave right now...

never had over a hundred spam emails before

I haven't used email for ~10 years. Recently, the email linked to Discord was deleted. Because I haven't visited it for a long time, I didn't even know what they were doing, but it turns out IT'S NORMAL ahahaha... The funniest thing is that you can't change your Discord email without confirming a message from the email already linked to it. I had to set up two-factor authentication on my device to have the possibility of entry. So that's how we live.

The fun funniest thing about all this is that Discord directly says that if the email is deleted, write to the email service support team and generally sort it out yourself, don't write to us. <:BlobYay:806132268186861619> And don't care that the phone number is linked, that two-factor authentication is linked, that I know the login and password for the account, none of this matters, you just won't be able to log into the account if Discord decide to check your email, ever.

That once happened to me. I just registered e-mail with the same name. Because it was completely deleted, e-mail provider just let me do it. That allowed me to keep using accounts on all sites that e-mail was linked to.

https://www.theguardian.com/world/2026/feb/08/memorial-for-swiss-bar-fire-victims-crans-montana-goes-up-in-flames

can we just stop causing fires

i wanted to get the new steam controller but just learned that it can only be used as mouse and keyboard on platforms that don't explicitly support it, not as an actual controller... seems completely unreasonable to me

more specifically, the original steam controller is in "lizard" mode (emulating mouse + keyboard) unless it periodically gets told to act as a controller and from the code that valve contributed to sdl3 that seems to be the same on the new controller that's coming out sometime this year

(and who thought it was a good idea to give these two completely different controllers the exact same name anyway)

the old Steam Controller can work as a regular controller on Linux without Steam

it does have a driver, but even if it does support the new controller (idk if it does) i assume most things like my phone and tv won't have it

can't find any reports of the old one working on an android device at least

that headline made me think it was The Onion

<@853026420792360980> if you don't mind, I have an mpv question. I found this snippet online for bringing up the FPS of a video into one's monitor's VRR range ``` video-sync=audio [vrr1x] profile-cond=get("container-fps", 999) >= 37 vf-remove=@vrr [vrr2x] profile-cond=get("container-fps", 999) < 37 and get("container-fps", 999) >= 25 vf-remove=@vrr vf-add=@vrr:fps=fps=2*source_fps [vrr3x] profile-cond=get("container-fps", 999) < 25 vf-remove=@vrr vf-add=@vrr:fps=fps=3*source_fps ``` But I just read in the manual that the container-fps property isn't very reliable. I've tested a few files and it seems to work. Do you reckon it'll be fine?

This is not something you should use anyway

regardless

the `fps` filter changes the framerate of the input video by duplicating or dropping source frames to match the desired framerate. for example, if your monitor's vertical refresh is 60 Hz, then this has the effect of taking 30 fps video and duplicating every frame to make it 60fps

this doesn't actually accomplish anything

if the vertical refresh rate is an integer factor or multiple of the framerate, then duplicating or dropping frames doesn't do anything interesting

My recommendation is to use `video-sync=display-resample`

this is usually done by the gpu (LFC) if the fps is below the VRR range but idk if/how doing it in software helps

I don't think my monitor has LFC. Not sure on the GPU driver.

nothing needs to be done in the integer case though

pretty sure both nvidia and amd (idk about intel) just always do it automatically

like, if you have a 60 Hz monitor displaying 30 fps video, then you only need to repaint 30 times a second

this is independent of GPU or driver. you don't need to duplicate frames

My monitor has a range of 48 to 75 Hz. I want my screen to be perfectly synced to the video.

is that not the case currently? like did you run something like a judder test video and see issues

You want `--video-sync=display-resample`, not `--video-sync=audio`

`video-sync=display-*` will sync the video frames to the monitor's refresh rate

`video-sync=display-resample` will time frames to the monitor's refrseh rate, and will slightly resample the audio to compensate for drift

this is typically what you want

But I don't want to sync the video to my monitors nominal refresh rate, I want variable refresh rate monitor to draw a new frame every time mpv / the GPU driver sends one over the wire.

Oh you are looking for `--drm-vrr-enabled=auto`

I think

do applications have to opt in to vrr on linux?

I don't know if gpu-next supports variable refresh rate yet. I know the DRM driver does.

you're still probably better off using `--video-sync=display-resample` since mpv already handles the duplication and dropping of frames and repaints. inserting a video filter to do it doesnt make any sense

there's minimal that you need to do beyond `video-sync=display-resample interpolation=yes tscale=catmull_rom` to have a good experience in that regard

My monitor's OSD is pretty bad at displaying the refresh rate, but at least I can tell that it sits at exactly 75 Hz when I use `video-sync=display-resample`, and that it doesn't when I use `=audio`.

So something is working right in the latter case, it's just hard to know exactly if all is as it should be.

Alright, I *think* it's working correctly with `video-sync=audio` without the filters. I didn't try it independently before because people were saying they needed to do that to get it to work, but OTOH that was a few years ago.

and yeah I'm running under Wayland so the DRM options aren't relevant to me. I think `--drm-vrr-enabled=auto` just does the same as what Sway does when you tell it to enable VRR anyhow.

what about mpv's OSD

supposedly we support it on gpu-next

What am I looking for here?

`i`

seems like it only gets the nominal refresh rate in audio sync mode.

In display-resample it also reports the estimated rate

check that

that just jitters around the nominal rate, presumably since that's what it's trying to match

hm

supposedly it does support vrr but idk much about it

I asked on irc and got "yes" but nobody else is available to answer

ah

my recommendation is to ask in #mpv on libera and wait

I tried looking for some way to get frame times from the DRM system, but I couldn't find anything.

This feels like it's up DMG or sfan5's alley

I know more about software stuff than hardware

mm

maybe kacper

I'll hop over there in a bit, though maybe tomorrow cuz I need to head to bed.

Here is what I can observe: when I run VRRTest and set the FPS to anything from 49 to 75 FPS, the monitor's OSD pretty consistently shows that FPS, with a little flickering to other values. When I play a 60 FPS video in mpv, the monitor OSD sorta flickers between 59 and 60, with the occasional jump to 75. When I play 24 FPS video (or 23.whatever), or set VRRTest to a similar low value, the monitor OSD goes crazy flying up and down between 48 and 75. Which tells me yeah the GPU isn't sending out new frames at the nominal rate, so VRR is working, but it doesn't really tell me if it's working correctly.

fwiw 24000/1001 is adjusted to 24

they pretend X/1001 is X/1000 and resample the audio accordingly

cause fuck that lmao

lol

I found that this just resamples to the "set refresh rate" and not any lower