jsonstein@masto.deoan.org ("Jeff Sonstein") wrote:
this suxs
h/t @StillIRise1963
jsonstein@masto.deoan.org ("Jeff Sonstein") wrote:
this suxs
h/t @StillIRise1963
Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
TheConversationUS@newsie.social ("The Conversation U.S.") wrote:
Medicaid enrollment surged during the pandemic, then tumbled during the ‘Great Unwinding’. A health economist who studies the effects of public policy on insurance coverage looks at how paperwork and particular state policies influence who is still covered.
https://theconversation.com/25-million-people-lost-medicaid-after-the-covid-19-pandemic-and-state-policies-shaped-who-stayed-covered-277599
Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
wdlindsy@toad.social ("William Lindsey :toad:") wrote:
"On Monday, a federal judge dismissed Trump’s $10 billion defamation lawsuit against the Wall Street Journal. The lawsuit alleged that the Wall Street Journal defamed Trump when it reported on a bawdy letter allegedly authored by Trump in 2003 in celebration of Jeffrey Epstein’s 50th birthday."
~ Judd Legum
Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
brucelawson@vivaldi.net ("Bruce Lawson ✅ ♫ ♿ ✌️♂️✊") wrote:
Man Who Threw Molotov Cocktail At Sam Altman’s Home Claims He Was Following ChatGPT Recipe For Risotto https://theonion.com/man-who-threw-molotov-cocktail-at-sam-altmans-home-claims-he-was-following-chatgpt-recipe-for-risotto/
Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
wdlindsy@toad.social ("William Lindsey :toad:") wrote:
"All over Fox News and social media, conservatives are absolutely losing their minds over birth rate dips, and are upset that teenage girls aren’t having babies like they used to. ...
When I was a kid in the 1990s, it was conservatives even more so than liberals who wrung their hands about 'babies having babies.'”
~ Jill Filipovic
#Republicans #MAGA #GreatReplacement #ReproductiveRates #BirthRates #babies #WhiteSupremacy #TeenMothers
/1https://www.throughline.news/p/want-more-babies-make-more-girlbosses
Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
wdlindsy@toad.social ("William Lindsey :toad:") wrote:
"Now the Pope is saying, Now let me tell you what the scriptures really say. And let me just meet you with the scripture. It's not my words. It's not my politics. It's not my policy. I'm just going to hit you with the scriptures. Blessed are the peacemakers, the children of God. Blessed are the meek."
"In a real sense, with Donald Trump, what you have at MAGA is a war on divinity."
~ William J. Barber
#Trump #PopeLeo #PaulaWhite #FranklinGraham #narcissism #WhiteChristianNationalism #MAGA
/5
Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
wdlindsy@toad.social ("William Lindsey :toad:") wrote:
"One thing that I love of what's happening right now is the Pope is not speaking as a political figure or as a politician. What he is doing is quoting the scriptures ans exposing these hustlers like Paula White and Franklin Graham and all the rest of them who have been hijacking the scriptures to pump this guy up and justify all their evil, their wickedness that they've been doing."
#Trump #PopeLeo #PaulaWhite #FranklinGraham #narcissism #WhiteChristianNationalism #MAGA
/3
aredridel@kolektiva.social ("Mx. Aria Stewart") wrote:
So to follow up on this, I've caught it in action. Models, when quantized a bit, just do a bit more poorly with short contexts. Even going from f32 (as trained) to bf16 (as usually run) to q8 tends to do okay for "normal" context windows. And q4 you start feeling like "this model is a little stupid and gets stuck sometimes” (it is! It's just that it's still mostly careening about in the space of "plausible" most of the time. Not good guesswork, but still in the zone)
And then at Q2 (2 bits per parameter) or Q1, the model falls apart completely. Parameters collapse to zero easily. You start seeing "all work and no play makes jack a dull boy” sorts of behavior, with intense and unscrutinized repetition, followed by a hard stop when it just stops working.
And quantization is a parameter that a model vendor can turn relatively easily. (they have to regenerate the model from the base with more quantization, but it's a data transformation on the order of running a terabyte through a straightforward and fast process, not like training).
If you have 1000 customers and enough equipment to handle the requests of 700, going from bf16 to q8 is a no-brainer. Suddenly you can handle the load and have a little spare capacity. They get worse results, probably pay the same per token (or they're on a subscription that hides the cost anyway so you are even freer to make trade-offs. There's a reason that subscription products are kinda poorly described.)
It's also possible for them to vary this across a day: use models during quieter periods? Maybe you get an instance running a bf16 quantization. If you use it during a high use period? You get a Q4 model.
Or intelligent routing is possible. No idea if anyone is doing this, but if they monitor what you send a bit, and you generally shoot for an expensive model for simple requests? They could totally substitute a highly quantized version of the model to answer the question.
There are •so many tricks• that can be pulled here. Some of them very reasonable to make, some of them treading into outright misleading or fraudulent, and it's weirdly hard to draw the line between them.
pzmyers@freethought.online ("pzmyers 🕷") wrote:
Paperwork submitted. I am officially unemployed as of April 2027.
https://freethoughtblogs.com/pharyngula/2026/04/14/its-official/
baldur@toot.cafe ("Baldur Bjarnason") wrote:
The slides for the talk with all of the references. Lots of worthwhile reading there
https://faculty.washington.edu/ebender/papers/Bender-ISU-2025.pdf
baldur@toot.cafe ("Baldur Bjarnason") wrote:
"Resisting Dehumanization in the Age of "AI": The View from the Humanities, Emily M. Bender"
https://www.youtube.com/watch?v=T7Lc6QNxolQ
> What if we had a bunch of money that came in from tobacco interests? Would we have a series of grants to explore how tobacco can enhance teaching and learning?
baldur@toot.cafe ("Baldur Bjarnason") wrote:
RE: https://dair-community.social/@emilymbender/116397605266279416
It's a bit of a trope to say "this is an important talk" when somebody tackles an sombre topic, but this is genuinely an important talk on the acceleration of dehumanisation enabled by "AI" and how we can resist it, all wrapped in a decent overview of the discourse so far, from a leading expert in the field
jsonstein@masto.deoan.org ("Jeff Sonstein") wrote:
this is most excellent, a sign of growing software maturity:
Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
benpate ("Ben Pate 🤘🏻") wrote:
Have you seen this news?
#Mastodon just got funding to add end to end encryption into their software.
So, some time next year, you’ll be able to send truly private messages to the vast majority of the #Fediverse
Im so excited about this.
Because it’s an open spec, this opens the doors for every Fediverse app to join the party.
Yesterday, this project was a proof of concept. Today, Mastodon has turned it into a stampede.
https://blog.joinmastodon.org/2026/04/sovereign-tech-agency-funding/
Boosted by jsonstein@masto.deoan.org ("Jeff Sonstein"):
MikeDunnAuthor@kolektiva.social wrote:
Today in Labor History April 14, 1816: Bussa led a slave rebellion in British-ruled Barbados. Because of this, he is remembered as the first national hero of Barbados.
#workingclass #LaborHistory #slavery #rebellion #racism #barbados #bussa #BlackMastodon
jsonstein@masto.deoan.org ("Jeff Sonstein") wrote:
Trump is claiming a "blockade" of the Strait of Hormuz, but this appears to be true only for some definitions of "blockade"... if your ship is connected to the PRC, for example, then there is apparently no "blockade" in place at all.
an attempt at a quiet "nod nod, wink wink" TACO move? or just, once again, action/reaction that was not actually thought through? I can no longer tell.
Boosted by GuillaumeL@hachyderm.io ("BigSaur G"):
paul_denton ("Nils Wilcke") wrote:
Macron isolé, la purge a débuté avant 2027: ses conseillers se pressent pour quitter l'Elysée et les téléphones sonnent dans le vide. Et une enquête sur la première année de mandat de Ferrand au Conseil constitutionnel. Le macroniste bouscule l'institution, au risque de l'affaiblir? https://www.youtube.com/watch?v=447Pq6JPPxw&t
#Politique #ConseilConstitutionnel #Macron #Compol #Justice #Droit #RH #Elysee #Presidentielle #Guerre #Eglise #Vatican
dysfun@treehouse.systems ("gaytabase") wrote:
okay bear with me here but if you get NaN, you done fucked something up already. personally i'd rather know instantly, but if you want your data to be silently corrupted, why not just use NaN as the null value?
dysfun@treehouse.systems ("gaytabase") wrote:
hmm i wonder how much people would hate it if NaN and null were interchangeable
adele@social.pollux.casa ("Adële 🐁") wrote:
What's new in #SmolFedi v1.0.11 the html/css/no-js fediverse #smolweb client :
- Register/Unregister to new posts on profile page
- Manage follow requests
- Add Favourites, Local and Federated timelines
- Manage post moderation : mute/unmute thread / report post
- Manage user moderation : mute/unmute block/unblock report
- Allow hr tag in html post (thx daj)
- Add poll posting
- Define session handler (compatibility)
- Add info 'follows you' in profile info page
- Permit post edition
- Manage word filters
- Simplify page width system
- Permit to mark media as sensitive
- Add search form
jscalzi@threads.net ("John Scalzi") wrote:
He's not the antichrist. The antichrist wouldn't be that stupid.
dysfun@treehouse.systems ("gaytabase") wrote:
does anyone have numbers on NaNboxing overhead?
ChrisWere@toot.wales ("Chris Were ⁂🐧🌱☕") wrote:
Me and Drew are playing a little Div on #Owncast
aredridel@kolektiva.social ("Mx. Aria Stewart") wrote:
RE: https://toot.cafe/@baldur/116402441445902779
This but also ratcheting up quantization to reduce cost.
db@social.lol ("David Bushell 🪿") wrote:
UK/EU web devs - have you noticed US business enquiries have gone quiet in recent years, or is it just me, did I scare them away?
We’re happy to share that Mastodon has been awarded a service agreement from the Sovereign Tech Fund @sovtechfund 🎉
This covers five major initiatives through 2026 and 2027. We are very grateful for this support. Read about the details in our blog post.
https://blog.joinmastodon.org/2026/04/sovereign-tech-agency-funding/
dysfun@treehouse.systems ("gaytabase") wrote:
RE: https://live.acarsdrama.com/@acarsdrama/116402682761183769
even pilots are fed up of javascript
dysfun@treehouse.systems ("gaytabase") wrote:
hypothetically, where would one start if one wanted to add a new index type to postgres?
dysfun@treehouse.systems ("gaytabase") wrote:
i kinda feel like i should learn enough verilog to design an actually good SIMD ISA
Boosted by dysfun@treehouse.systems ("gaytabase"):
Baa@mk.absturztau.be ("Baa 🐑") wrote: