Hi everyone this is Lee I’m trying mastodon again :)
@butchanarchist welcome to kolektiva's server!
@somebody but again duda himself seems to get this shit. from the tANS probability distribution paper, final page:
Generally finding the optimal symbol spread seems to require exponential cost, but such iterative tuning, or maybe just searching permutations of neighboring symbols, could lead to some improvements - practical especially for optimizations of tANS automata for fixed distributions.
that is exactly what i think the answer is. an exponential algorithm that tries merging and splitting every single symbol that occurs in sequence together
@somebody one really curious little tidbit about discrete sequential coding like this is that contiguous ranges of k tokens in the input stream correspond to a kth derivative
Claudius Link
boosted
RE: https://digitalcourage.social/@BlumeEvolution/117197560205910448
Guten Morgen - Tässle Kaffee ☕️?
Worüber ich mich am meisten geirrt habe?
Als junger Mensch glaubte ich, dass es für Wissenschaftler immer gut ist, wenn sich Thesen bestätigen.
Doch in Wirklichkeit fühlt es sich oft mies an. Ob Klimakrise oder Ressourcenfluch - oft schmerzen schon die Prognosen, dann ihr Eintreten.
Und: Menschen mögen keine Rechthaber. „Wir haben es Euch doch gesagt!“ löst neue Reaktanz aus & befeuert Verschwörungsmythen. Und auch umgekehrt: „There is no Glory in Prevention.“
Heute weiß ich: Die Evolution hat die menschliche Psychologie nicht auf abstrakte Wahrheit, sondern auf Zugehörigkeit geprägt. Wenn Du mehr als nur Deine eigene Blase bespielen willst, so musst Du den schweren Weg wählen: Dialog. (Philosophisch: Dialogischer Monismus.)
Und das mache ich jetzt seit Jahren auch im Fediverse. Allen, die sich ehrlich interessieren: Vielen Dank vom #Solarpunk 🙏🤔📚🔋
mkb
boosted
Has anyone considered that mind is the fear killer?
Lainey Lou the Doodle - https://runs.lol/2026/09/01/lainey-lou-the-doodle
4 media files
@ericmwalk Aww, I'm so sorry. She was clearly very loved!
So tragic what we are doing to people
What We Did to This 20-Year-Old Is Unforgivable
YouTube
Two versions of a graphic. Feel free to say which you prefer. And if you are an actual graphic designer who would be willing to improve one or both, pro bono, do DM me.
2 media files
Claudius Link
boosted
Feeling the post #Socrates blues? Jealous of the smart conversations you're missing? Eager to win a free ticket to BrewtCon, an actual in-person open space?
Then come to FroGS on Saturday! We'll make you forget you ever had a Deutsche Bahn DeBacle™️ and make joining a video call on a Saturday worth your while.
🐸 September 5, 13:00 CEST https://frogsconf.nl/ #FroGSConf
@somebody however i unfortunately have kind of a negative view of the probabilistic assumptions employed. i think duda is actually a genuinely good scientist and is doing his best but there's so much more depth than this.
to wit, from the pyramid vector quantizer paper he cites in the paper on generating probability distributions for tANS:
In the case of the mean-square error performance criterion, a construction analogous to 1)-3) (using (9) for X and selecting the representation regions as spheres instead of pyramids) can be shown to provide asymptotically (in L) optimum performance in the small distortion case.
For either performance measure, however, the construction is based on a random assignment of points to S(L, L/λ) and leads to designs which are extremely difficult to implement.
and they do a whole bunch of gaussian bullshit which i just hate
Cookie monster! (PDF)
1 more link(s)
@somebody but again duda himself seems to get this shit. from the tANS probability distribution paper, final page:
Generally finding the optimal symbol spread seems to require exponential cost, but such iterative tuning, or maybe just searching permutations of neighboring symbols, could lead to some improvements - practical especially for optimizations of tANS automata for fixed distributions.
that is exactly what i think the answer is. an exponential algorithm that tries merging and splitting every single symbol that occurs in sequence together
hypebot
boosted
In today's picture: Corvus frugilegus (known as rook). #CORVUSPHERE #CorvusFrugilegus #Rook 
🤔 Could AI systems ever be welfare subjects?
🎥 New recording! As part of our AI Sentience Scholars program, Winnie Street and Geoff Keeling (Google Research) explore "Emerging Questions in AI Welfare"
First public release of Uxn68k
Still very new software, might not work at all on your machine. (But if you do try it out let me know how it goes!)
4 media files
@d6 So many memories. BBEdit was the first editor to truly bring me joy. When the BMUG ResEdit book came out, I had a blast with it.
Mre. Dartigen [renovator mode]
boosted
I updated HowSketchy to version 1.0 today! I think it works properly overall and the database is slowly getting some volume.
I could still use help with the latter though. I think it will improve both quality and quantity if there's more than one maintainer 😊
Mre. Dartigen [renovator mode]
boosted
Okay, let's go public. 😊
HowSketchy provides an overview of bands and artists, documenting problematic behavior as well as notable merits where supported by reliable sources~
It's not limited by genre or style.
So if you want to see which artists are safe or which ones to avoid, have a look:
The database is still very small, but will hopefully grow quickly.
If you'd like to contribute making this platform better, you can become an editor:
🌍 Help us translate #PeerTube!
The software is currently partially translated into many languages, but unfortunately, it is only 100% translated into English.
If you have language skills, you can help here: https://weblate.framasoft.org/projects/peertube/angular/
Thanks a lot! 💖
@peertube The English translation needs some important polish in key places. Where can we make suggestions?
Claudius Link
boosted
I don't really celebrate my birthday, but that doesn't mean you can't celebrate it for me! Donate to your favorite anti-genAI open source project, finally commission that artist you've been eyeing, give money to a queer/trans person that's struggling rn, and/or punch a Nazi for me. Give yourself permission for self-care. Live authentically. Just generally and genuinely be a cool and good person. We need as many of those as we can get especially now.
Comenzó super el miércoles. Y uds mi tuntunes?
Claudius Link
boosted
One thing that keeps me up at night is the progressive certainty that we as a society are slowly but surely losing our ability to be bad at things. We are ceding that skill to AI.
"You don't understand, without AI, I couldn't have drawn this/written this/designed this." says a person, certain that surely this, this argument right here is the nail in the coffin of the anti-AI argument. Utility: "Without this thing, I would have been bad at this task."
Yes. Correct. You would have been bad at it. And that's ok. Not only because it's just... Ok to be bad at things in general. But also because being bad at a thing is the first necessary step to learning how to do a thing.
Every musician you love has spent thousands of hours first being bad at being a musician. Every artist that brings you to tears has spent thousands of hours being bad at art. Every single designer with a master's degree in design has spent thousands upon thousands of hours designing things that are utter crap.
We learn how to do a thing by first being bad at it. And then noticing what skills we need to acquire to be better at it. And then honing those skills on a million failed attempts. And THEN we are finally good at doing it.
And being bad at things isn't just about learning how to do it yourself. It also teaches you how to listen more effectively to music. How to see art with the eyes of an artist. How to study a design for design flaws before it's even built. It teaches you good taste. How to perceive the humanity in the things all around you - things built by humans, for other humans.
No, the biggest problem with AI isn't that it makes bad art - although it most assuredly does also do that. It's that it robs human beings of the legitimacy of creating bad art ourselves.
How can anyone ever be expected to learn to do a thing if we can't ever be bad at it first?
Air France sends me an email to advertise its high-speed wi-fi on board... (note, it didn't say "free").
Dudes, I hate flying, but one of the only things I like about flying is that for a few hours, you can finally be disconnected with no way to give in and go online.
