Skip to main content




Lotus still knows how to make a driver’s car: The 2025 Emira V6, driven
Forget the Chinese-made Lotus EVs, its UK-made mid-engine sports car still rules.
arstechnica.com/cars/2025/08/l…
in reply to Ars Technica

Vous voulez pas acheter un vélo plutôt !?
Nan, j'dis ça, c'est juste pour retarder qu'on crame ensemble au volant de superbolides au milieu des megafeux (ou dans le flot d'inondations monstre) ...
Mais c'est vous qui voyez, hein 😜
in reply to Ars Technica

Could you please stop hyping gas guzzling anachronisms?
This entry was edited (4 months ago)


LETO DNI AKTA O UI: ZAKAJ SO NAŠE PRAVICE ŠE VEDNO NA UDARU?

Danes mineva eno leto, odkar je akt o umetni inteligenci postal zakon EU.

A v času, ko se pripravlja njegova polna uveljavitev, si EU in nekatere države članice že prizadevajo rahljati težko priborjene zaščitne ukrepe.

Kaj je na kocki in kako lahko obranimo svoje pravice pred represijo in nadzorom s podporo umetne inteligence? ➡️ danesjenovdan.si/leto-dni-akta…






Scene at a Safeway in Windermere, Edmonton, after a group of four people filled their carts with what staff estimate was at least $1,500 in groceries and left without paying.

They look like Gyppos to me.





TO BILK. To cheat. Let us bilk the rattling cove; let us cheat the hackney coachman of his fare. Cant. Bilking a coachman, a box-keeper, and a poor whore, were formerly, among men of the town, thought gallant actions.

A selection from Francis Grose’s “Dictionary Of The Vulgar Tongue” (1785)

--
#books #literature #dictionaries #history #society #language #slang @histodons

histodons group reshared this.








Running Deepseek R1 671b fully locally on a $2000 EPYC server. Idle wattage is just 60w while with 260w under load.

This setup runs a 671B model in Q4 quantization at 3-4 TPS, running a Q8 would need something beefier. To run a 671B model in the original Q8 at 6-8 TPS you'd need a dual socket EPYC server motherboard with 768GB of RAM.

The idea that LLMs use an inordinate amount of power to run is very much outdated at this point.

digitalspaceport.com/how-to-ru…

#technology #llm

This entry was edited (4 months ago)

#tech reshared this.

in reply to 🦃 Ins0mniak 🦃

@ins0mniak @adiz I run them on M1, but there's a definite difference in quality of output between smaller models and a full sized one
in reply to 🦃 Ins0mniak 🦃

thats nuts, i thought LLMs required tons of VRAM? i'm definitely missing something here, what is it?
in reply to Vinnewed Goblin

Hardware has gotten a lot better/optimized for LLM (e.g. Mac M-series chips have dedicated LLM units) + models have become a lot more optimized to where you can get very performant and useful models that are small in size and execute efficiently. @ins0mniak @yogthos
in reply to 👺кину奇诺[流浪者]👹

@warmbeverageenjoyer@fsebugoutzone.org @ins0mniak yeah this stuff is moving at insane speed, I fully expect that we'll be able to run what's equivalent of top models today on consumer grade hardware in a coupe of years
in reply to 👺кину奇诺[流浪者]👹

that's super cool, and i didn't know that about the dedicated LLM units
i try to avoid using LLMs (i fail sometimes) but i'd probably prod a completely local model i picked myself much more often
in reply to Vinnewed Goblin

Ive been using mine to play rpg games. Its actually pretty good with it.
in reply to 🦃 Ins0mniak 🦃

this is my #1 use case for them, but the free online ones are pretty shit at remembering anything that happened more than 3 messages ago and also are a privacy nightmare
when i tried to mess with them on my desktop pc like a year-ish ago, i couldn't get anything remotely worthwhile going on my 8gb 3060ti lol.
what sort of model(s) are you able to run?
in reply to Vinnewed Goblin

So far Ive run a few of the gemma ones and deepseek via Ollama.

All perform much better than the intel Orin development board I got. The only really issue Ive had with some of them is diskspace but other wise they run as fast as any online one.

in reply to Vinnewed Goblin

Its a lot of fun messing with it. I love the idea of being able to run them locally


Oof!

Keep 'em coming

Via Kyle Cheney:

BREAKING: Judge Cobb has blocked the Trump administration's use of summary deportation for those who had previously been paroled into the U.S. — and warns the 👉🏼tactics used by the administration are beginning to resemble the countries these people fled.👈🏼 ecf.dcd.uscourts.gov/cgi-bin/s…



Aircraft from Jordan, the United Arab Emirates, and, for the first time in over a year, Spain, Germany, and France are airdropping humanitarian aid packages over the Gaza Strip today.

Footage shows the airdrop over central Gaza this morning.
#AureFreePress #News #press #headline #Israel #gaza #Breaking #BreakingNews