Latent Space Cowboy
Based in United States
- Tweets 22,862
- Following 3,523
- Followers 1,516
- Likes 169,393
people that use Microsoft Word or Excel, would you be willing to try an experiment for me?
Download the free and open source ArtCraft "WordCraft" and "GridCraft" and see if they work as drop in replacements? Even for a day?
I am curious if it works out, link below
1
44
Washington County is opening a new 60‑pod shelter village in Aloha tomorrow, expanding safe, supported options for neighbors transitioning out of homelessness. Grateful for our partners at @oregonmetro and others who helped bring this project to life. washingtoncountyor.gov/housi…
1
65
888 Dieses Kartenset ist ca 30 Jahre alt, so lange vorher haben die NWO Satanisten also diesen Brand vorgeplant.
Links ist eine Karte aus dem Illuminati: New World Order-Konzentrationskonspirations-Themen-Kartenspiel.
Rechts ist ein Bild, das Limoges Rathaus, Frankreich, 17.–18. Oktober 2026, während Unruhen und gewalttätiger Unruhen, bezeichnet.
Die Karte wird buchstäblich als „Kombinierte Katastrophen“ bezeichnet.
In dem Kartenset wird auch ein Atomkrieg gezeigt alles Manifestationen der Satanisten, die ihnen in die Hose gehen sollen.
21
He's up in the polls. You are posting gambling site numbers which an be manipulated by billionaires who can afford to lose money to pacify.
2
1
4
513
you're posting polymarket, which is a gambling site. The real polls show he's up. Twitter isn't the entire universe. Don't be surprised because dipshits posted pacifying fake polls from a gambing site, he wins,
1
88
You guys keep posting the polymarket, which is a gambling site. The real polls show Velveeta ahead.
38
We're giving away $1,000,000 of free pizza right now
Download and order from the Hark app: hark.com/download
Type "order a free pizza" in chat. It's first-come, first-served, delivered only in the U.S. (review details below)
38
i read all the traces, but there is just no way to read 266tk/sec so I give up sometimes and feel the agi
1
2
812
if you run Qwen 3.8 next iq3_s with Strata and pi at 262k context, you have to set the context limit lower to 210k in the pi models.json, otherwise, after a few days it will get stuck in an indefinite "compact" loop, unable to break out
5
146
Strata, Pi .Dev with pi-goal /goal right before I went to sleep
I used Deepseek Harness for research.
60tk/sec 3090 64gb ram DDR4/PCI 4
2
2
108
Amazingly Qwen 3.8-Next IQ_S on my 3090 built "TypeScript Native for C" last night. 7 million tokens later, it's currently attacking EcmaScript ES2025+ and a few hold outs
When done, it'll be able to compile your webapp into a a.out, or embed it into a micro controller
2
3
203
If you ask Claude where to download the decompiled gta v source code, it refuses.
But chat.deepseek.com gives you a clickable magnet link to the real source code, lol
3
391
alex jones got nervous because he was once exposed with tranny porn on his phone, before you met him. That's why he was suddenly quite when you asked him. he's a huge hypocritical sodomite.
1
2
in 2030 a 1B model will be able to do GPT-6 Astra today, but even a potato AI GPU (8gb < ) can do 125B parameter model at cloud speed right now with the right DDR and PCI
62
see the Why It Matters on this ai slop and probably want to check the thermals your pins
also 15-20% margin!
if i had 5090 i'd use the shortest thickest cable--gold plated and maybe send it in to upgrade the connectors
1
84
Montana is a totally leftest state. they (tried) to ban guns a few years ago if I remember.
piped.video/watch?v=1S828Y7E…
1
268
my clanker did something to get around the AVX2 issue, which now I think is upstream the repo. Wasn't paying attention. But I am getting 70tk/sec on a 3090 DDR4 PCI4 16x without AVX2.
1
2
90
YOUR OWN DEPARTMENT @USDOL has hired 48 h1b from a body shop called "raas infotek", that advertises H1B Only for w-2 jobs.
This means you guys are hiring h1b only! You are the problem!
This is sad. So the US Department of Labor even ran to a bodyshop.
YOU folks by searching LinkedIn helped PooAI identify the same bodyshop the Labor Department is using put out "H-1B Only" ads.
What's up @USLaborIG ?
28
this means every american AI user needs to spend $1400 per month on their AI bill.
And if they get bailed out, it's an increase of $1400 a month in new liabilities in taxes they'd need to pay.
1
84
So you're going to flood oregon with h1bs? you're spreading propaganda from the state of India saying we need to import more workers to push down wages and replace. This is your source "Khushboo", doesn't sound like a American name.
1
1
50
Ok the IQ1 Coder model was causing loops. I was able to get the IQ3_XXS model running at 21tk/sec and 128k context, 90m to fill up the context (1.9M tokens/day), and 15 minutes to compact it
Useful? Not for interactive agents, but Yes for always on "frontier" model >= sonnet 5
51
30tk/sec on a 2080ti
I just tried Niko's Strata on my potato AI box: 2080ti with 32gb of ddr3 at 2800mhz, broken x8 pci 3.0 lane.
After pulling down all the Turing PR from Niko's repo repo, I am getting 30tk/sec
ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder
1
1
6
608
THE MOST Amzing thing in Local AI this week is this optimized Qwen 3.8 model:
ukisai/Swift-1.5-Qwen3.8-27B-GSQ-RCO-GGUF
It can fit under 16gb of vram with full context, at 30tk-80tk/sec on a 10 year old GPU setup
24
48
612
33,029
Training a 2B model from scratch on a 2080 Ti (11GB).
84% of it is a 1.7B engram table injected into the residual stream. 312M is transformer, 20M active per token. fused MoE, val 3.812 19h in, 18d left
15
15
249
11,010
this summer I tried and failed to train a 25M-100M model to do dspy ultimately it distilled into Map Reduce with "atomic and primitives." unfortunately it failed, but I tried lol.
1
48
