I use Quad9 where I can, but admittedly also have some reliability issues. Their local PoP will drop all requests for a couple minutes at least a few times a week. It's a shame since it's the closest/best recursive resolver otherwise. CloudFlare is similarly close, but they don't support EDNS, so I've had problems because of that.
Communities seem to appear basically like a user, and a post in a community looks like a "boost" from that user.
if you look up @gamedev@lemmy.world on your instance that's where "we" are.
The vast majority of these 3d printed guns use the fact that the registered "gun" is only a small part of the actual whole gun. You can buy barrels, magazines, slides, firing pins, etc, all mail order without any fuss — it's the part with the serial number that is actually the firearm and restricted. So if you print that part (which is often basically just a shell to hold the metal components in place) you have an unserialized gun that's basically just as good as one from a gun store.
Cap LLMs to 1 Trillion parameters, can utilize no more than 384GiB of (V)RAM.
Offenders go to jail (corporations charged as a criminal enterprise), all equipment seized and destroyed.
If these AI glazers really believed that AI is the threat they claim it is, then we should treat it the way we'd treat corporations if they tried developing their own nuclear weapons.
The article is sort of sparse on details, are they directly rasterizing these textures from the vertex shader (or custom compute shader rasterizing)? Or just rasterizing to image buffers and sampling those like normal textures?
If the former, seems like you could do some neat stuff with LOD by removing/retopo-ing/recoloring the vector shapes as they get further from the camera.
Probably in anticipation of some serious price increases. They'd still like to get devices in people's hands even if they can't nominally afford them anymore.
It sort of depends on what you mean by can't reproduce without looking at hints.
Like, even now if you asked me to write a number guessing game in a language I've had to use a lot (Java) I couldn't do it straight from memory. I haven't needed to read stdin manually for a long time so I'd have to look at the class and docs to see the best way to pull in a line and parse it into a number, I don't remember the specifics of the Random class either, so I'd need to double check the methods available and if they produce inclusive or exclusive ranges.
To me that's normal; I understand the program flow and what needs to be accomplished, but the specific classes and methods are something I don't keep with detail in my head, using the docs and keeping the language reference handy is normal, in my experience.
Double sucks because it's also being used as a proxy flamewar between Zig and Rust enthusiasts.
Which is awful because there are good conversations and teachings to be had between those languages; but instead all of the circumstance is just blood in the water and completely unproductive.
Where’s the pressure? Who’s making these files and trying to use them?
The standards bodies and the companies that pay their bills. That's Netflix, Google, et al. All have a vested interest in good multimedia support in browsers. It's why browsers have shockingly good support actually, most of the problems are downstream on clients and implementations that don't know or care to enable these newer formats.
You decide what to target, and work with what you have. That's how it's always been with web technology.
If you need to support the barest minimal possible clients then you might not even have HTML5. Otherwise, if you have HTML5 you can use the source tag to provide clients with a choice of formats/encodings and they can pick the one they support. If the clients can't even mange that, well, things on the internet have always been broken anyway.
And somebody's gotta start somewhere, WebP was shoved down everyones' throats and now it's supported very broadly, as AVIF pick up users software devs will take notice and start supporting it.
It's not the route I'd take, after all, the video tag in html supports specifying multiple sources (including different formats/encodings), and with full control of the stack (they vend Safari) you could have logic on devices without AV1 hardware that prefers the AVC/HEVC/VP9 sources instead — then fallback to AV1 SW if it's the only option. That seems a better user experience than just failing to display content.
I checked that app's source code, and basically they're doing their own avif decode call instead of relying on a browser/webview to handle it. So essentially they forgot/don't know/don't intend to implement animated avif. Looks like they also use the aom reference decoder instead of using something faster like dav1d.
It's an intentional behavior by Apple. Basically they just don't support AV1 videostreams unless the hardware you're using has a hardware decoder (read: very new). They could support it using software decode (what browsers typically do for AV1 inside avif containers) but.... for whatever reasons don't.
Apple has limited support for AV1 streams (yes, even for software decode) unless on very recent hardware.
Here's an AV1 stream inside a webm container for comparison, would be interesting to see if that works over the avif container on your stack.
I use Quad9 where I can, but admittedly also have some reliability issues. Their local PoP will drop all requests for a couple minutes at least a few times a week. It's a shame since it's the closest/best recursive resolver otherwise. CloudFlare is similarly close, but they don't support EDNS, so I've had problems because of that.
Communities seem to appear basically like a user, and a post in a community looks like a "boost" from that user. if you look up
@gamedev@lemmy.worldon your instance that's where "we" are.The vast majority of these 3d printed guns use the fact that the registered "gun" is only a small part of the actual whole gun. You can buy barrels, magazines, slides, firing pins, etc, all mail order without any fuss — it's the part with the serial number that is actually the firearm and restricted. So if you print that part (which is often basically just a shell to hold the metal components in place) you have an unserialized gun that's basically just as good as one from a gun store.
Bro even Macy's sells these things in store. They're a mass market product.
Cap LLMs to 1 Trillion parameters, can utilize no more than 384GiB of (V)RAM.
Offenders go to jail (corporations charged as a criminal enterprise), all equipment seized and destroyed.
If these AI glazers really believed that AI is the threat they claim it is, then we should treat it the way we'd treat corporations if they tried developing their own nuclear weapons.
The article is sort of sparse on details, are they directly rasterizing these textures from the vertex shader (or custom compute shader rasterizing)? Or just rasterizing to image buffers and sampling those like normal textures?
If the former, seems like you could do some neat stuff with LOD by removing/retopo-ing/recoloring the vector shapes as they get further from the camera.
Probably in anticipation of some serious price increases. They'd still like to get devices in people's hands even if they can't nominally afford them anymore.
Also worth noting that since this includes coastline, it means basically 2/3 of Americans are deprived constitutional rights.
https://en.wikipedia.org/wiki/Border_search_exception#/map/0
They're working on the IPv6 part right now, just don't ask when it'll be done.
It sort of depends on what you mean by can't reproduce without looking at hints.
Like, even now if you asked me to write a number guessing game in a language I've had to use a lot (Java) I couldn't do it straight from memory. I haven't needed to read stdin manually for a long time so I'd have to look at the class and docs to see the best way to pull in a line and parse it into a number, I don't remember the specifics of the Random class either, so I'd need to double check the methods available and if they produce inclusive or exclusive ranges.
To me that's normal; I understand the program flow and what needs to be accomplished, but the specific classes and methods are something I don't keep with detail in my head, using the docs and keeping the language reference handy is normal, in my experience.
Double sucks because it's also being used as a proxy flamewar between Zig and Rust enthusiasts.
Which is awful because there are good conversations and teachings to be had between those languages; but instead all of the circumstance is just blood in the water and completely unproductive.
The standards bodies and the companies that pay their bills. That's Netflix, Google, et al. All have a vested interest in good multimedia support in browsers. It's why browsers have shockingly good support actually, most of the problems are downstream on clients and implementations that don't know or care to enable these newer formats.
You decide what to target, and work with what you have. That's how it's always been with web technology.
If you need to support the barest minimal possible clients then you might not even have HTML5. Otherwise, if you have HTML5 you can use the
sourcetag to provide clients with a choice of formats/encodings and they can pick the one they support. If the clients can't even mange that, well, things on the internet have always been broken anyway.And somebody's gotta start somewhere, WebP was shoved down everyones' throats and now it's supported very broadly, as AVIF pick up users software devs will take notice and start supporting it.
Using a browser or third party app?
I think that's a reasonable way to look at it.
It's not the route I'd take, after all, the video tag in html supports specifying multiple sources (including different formats/encodings), and with full control of the stack (they vend Safari) you could have logic on devices without AV1 hardware that prefers the AVC/HEVC/VP9 sources instead — then fallback to AV1 SW if it's the only option. That seems a better user experience than just failing to display content.
Very interesting, thank you!
I checked that app's source code, and basically they're doing their own avif decode call instead of relying on a browser/webview to handle it. So essentially they forgot/don't know/don't intend to implement animated avif. Looks like they also use the aom reference decoder instead of using something faster like dav1d.
That's caused by bad regex in the app, it's getting confused about domains.
e: attempted fix by using an aliased domain.
It's an intentional behavior by Apple. Basically they just don't support AV1 videostreams unless the hardware you're using has a hardware decoder (read: very new). They could support it using software decode (what browsers typically do for AV1 inside avif containers) but.... for whatever reasons don't.
Apple has limited support for AV1 streams (yes, even for software decode) unless on very recent hardware. Here's an AV1 stream inside a webm container for comparison, would be interesting to see if that works over the avif container on your stack.