Notices by Coyote (coyote@social.singing.dog)

@Mr_NutterButter @matrix I've always felt 343's halos looked off. Everything looks over designed and too busy, especially forerunner things. Speaking of, they really like to use floating things for forerunner stuff, but I hardly remember if any of that showing up in the original 3 games.

@BroDrillard @PurpCat @jimmybuffettfanaccount @Vidmastereon To be a good critic, you need to be able to identify and articulate real problems, but to be a good artist, you need to be able to understand and implement real solutions, and the greatest of both need at least the first part of the other. You can find this pattern everywhere as with the arrogance that being good at one side implies being good at the other ("This is terrible. I could make a better ~." "Don't talk to me until you've made your own ~.").

@SockPuppet @MK2boogaloo Enochian

@jeff @istvan @sun If it is then it's the long tail dragging it out. There's tons of teens and early twenties around that I'd believe you if you told me the median age was in the lower 20s.

@Inginsub @PurpCat @RustyCrab @vicious @swurl There’s a central server, the BGS, that collects content from instances, PDS, and passes it along to other servers to curate for those instances. Theoretically, you could host your own BGS, but you’d have to convince other people to connect to it. From what I understand, it’s like a mandatory relay.

@Inginsub @PurpCat @RustyCrab @swurl @vicious This was the blog where they go over their federation architecture and where I got that image:

https://blueskyweb.xyz/blog/5-5-2023-federation-architecture

@mint @PurpCat I find it pretty funny that AI research already thought of this idea to automate adversary to build more powerful AIs (generative adversarial networks). People praising these poisoning techniques as a means to fight AI assume newer AIs will never come out and are unaware that their efforts are creating more robust ones.

@RustyCrab I think they’re neat.

@RustyCrab @vriska @lanodan @coolboymew @goatmeal You can find this kind of stuff in places that exploded in population in less than 40 years. People build their houses in unincorporated areas where zoning is lax only for the city to swallow the surrounding land in development. The house never got sold or put up on the open market, so it never gets rezoned.

@xianc78 @beardalaxy @matrix @sjw @shibao Someone made a video of the fallout that spawned from that.

@mint

The value of owning the ecosystem cannot be overstated. Google itself has successfully used this paradigm in its open source offerings, like Chrome and Android. By owning the platform where innovation happens, Google cements itself as a thought leader and direction-setter, earning the ability to shape the narrative on ideas that are larger than itself.

Embrace, Extend, Extinguish.

@Flaky @PhenomX6 I’ve been blocked for “federates with the usual suspects” which, sure, is technically true, but is kind of stupid when the most I’ve done to their instance is follow an account. Also, they use masto.host which will firewall you randomly.

@PhenomX6 playing Eroge for PC-98 at the local coffee shop

@arcanicanis I’m not afraid that there won’t be places where it’ll exist, but that those places won’t reach the critical mass to be general purpose; If there’s only one place to post things touched by AI then that place will become that instead of the place that merely allows that.

SoFurry fell to the AI scare and placed a total ban on AI art, and here I thought they were playing it smart by just not addressing it. Of course, the only visible change was that a dude had to strip out all the illustrations from his story and a popular AI account that posted like once a week is going away.

@beardalaxy Have you met our lord and savior ffmpeg? Codec choice and bitrate are the most important aspects to consider when encoding audio or video. Newer codecs (such as AV1, H265, VP9, etc. for video) generally have better performance in quality per bit but have smaller hardware decoder support which could make the video choppy or unplayable on low performance devices. When targeting specific file sizes, it’s best to encode with a target bitrate using two-pass encoding instead of a target quality.

There’s also the question of quality vs size. You can encode a feature length movie into 16 megabytes, but it’s debatable if a movie of that quality would even be enjoyable to watch.

This calculator is useful for getting ballpark numbers on what bitrate to choose for video/audio files of a specific length to get a specific file size. If you’re working with a hard limit you might have to play around with them and other settings (framerate, resolution, audio channels, etc.) to squeeze the most out of it.

The attached file is encoded in VP9 and Opus in a webm container. It was encoded with a target of 16mb and overshot it by 0.2mb, and the frame rate had to be sacrificed, dropping down to 10 fps from 23.98. The total bitrate came out to be 27 kb/s (I think the audio was targeting 7 kb/s, so the video is around 18 kb/s), much less than the 1181 kb/s (972 + 195 kb/s) of the mpeg4 and mp3 avi original.

@PhenomX6

This is going to be hard to classify. Made me chuckle.

@Moon @tard @pawlicker @crunklord420 @RustyCrab All YouTube seems to recommend now are videos from the creator you’re watching and maybe one or two of the top trending videos at the moment. It honestly wouldn’t be any worse on a decentralized network where each node is a single channel. In fact, it’d probably be better if the system in place had things like video replies or community forums where people can share channels via word-of-mouth.

@parker @hermit @NEETzsche @bot There’s utility to the number-in-number-out of common ANNs beyond just being easier to train. While ignoring the temporal aspect is probably less efficient, it means you’ll get an answer (propagation of values through the network) in a fixed amount of time after training is complete which is valuable for the AI-as-a-tool use case.

As for whether or not it can host “true” artificial intelligence, I think any model that’s Turing complete has the capability to do it. The common ANN on its own can’t, but if one allows the network to run until it says it’s done and hooks up some of its untrained output into its input such that the system has a self regulated memory, it could develop some sort of self identity or conscious. Of course, we haven’t developed the math to train such an iterative network, so we’d be stuck with evolutionary methods.